Logistics track optimization method, device and electronic equipment for industrial chain

Abstract

Embodiments of the present disclosure disclose a logistics track optimization method, device and electronic equipment for an industrial chain. A specific embodiment of the method includes: conducting a no-fly zone detection on a flight area of a flight logistics device; obtaining a flight state vector sequence; in response to a flight remaining range being greater than a preset distance value, performing the following track optimization steps: determining a predicted flight state feature based on a track optimization model and a no-fly zone information set; inputting the predicted flight state feature into a neural ordinary differential equation network to obtain an optimized control parameter; controlling the flight logistics device according to the optimized control parameter, and determining a predicted flight state vector based on a flight dynamics equation; generating an updated flight state vector sequence according to the predicted flight state vector; and performing the track optimization steps again with the updated flight state vector sequence as the flight state vector sequence. The embodiment can reduce logistics transportation time and flight energy consumption.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular, to a logistics track optimization method and device for an industrial chain and an electronic device. BACKGROUND

[0002] In the flight process of a logistics transport aircraft, the no-fly zone avoidance problem is one of the core challenges in flight path planning. Currently, when optimizing the flight track, the commonly used method is a numerical method, which calculates the flight track by solving flight dynamics equations.

[0003] However, when the flight track is optimized by the above method, a technical problem often exists, which is that the numerical method usually requires complex integral operations, and relies on artificial experience to design a flight corridor, which consumes a lot of manpower and computing resources. In addition, due to the lack of a dynamic adjustment mechanism, it is difficult to achieve global optimal trajectory planning when bypassing the no-fly zone, which leads to an increase in transport time and thus increases flight energy consumption.

[0004] The above information disclosed in this BACKGROUND section is only for the purpose of enhancing the understanding of the background of the present disclosure and, therefore, can include information that does not form the prior art that is already known in this country to those ordinary skilled in the art. SUMMARY

[0005] The summary of the present disclosure is intended to introduce the concepts of the present disclosure in a simplified form, which will be described in detail in the following detailed description. The summary of the present disclosure is not intended to identify key or essential features of the claimed technology nor is it intended to be used to limit the scope of the claimed technology.

[0006] Some embodiments of the present disclosure propose a logistics track optimization method and device for an industrial chain and an electronic device to solve one or more of the technical problems mentioned in the background section.

[0007] In a first aspect, some embodiments of the present disclosure provide a logistics trajectory optimization method for an industrial chain, the method comprising: detecting a flight restricted area of a flight logistics device to obtain a flight restricted area information set; obtaining a flight state vector sequence of the flight logistics device; in response to a flight remaining distance corresponding to the flight logistics device being greater than a preset distance value, performing the following trajectory optimization steps: determining a predicted flight state feature corresponding to the flight state vector sequence based on a trajectory optimization model and the flight restricted area information set, wherein the trajectory optimization model comprises a long short-term memory network, a fully connected layer, and a neural ordinary differential equation network; inputting the predicted flight state feature into the neural ordinary differential equation network to obtain an optimized control parameter; controlling the flight logistics device according to the optimized control parameter, and determining a predicted flight state vector corresponding to the optimized control parameter based on a preset flight dynamics equation; generating an updated flight state vector sequence according to the predicted flight state vector and the flight state vector sequence; and performing the trajectory optimization steps again with the updated flight state vector sequence as the flight state vector sequence.

[0008] In a second aspect, some embodiments of the present disclosure provide a logistics trajectory optimization device for an industrial chain, the device comprising: a flight restricted area detection unit configured to detect a flight restricted area of a flight logistics device to obtain a flight restricted area information set; an obtaining unit configured to obtain a flight state vector sequence of the flight logistics device; and an execution unit configured to, in response to a flight remaining distance corresponding to the flight logistics device being greater than a preset distance value, perform the following trajectory optimization steps: determine a predicted flight state feature corresponding to the flight state vector sequence based on a trajectory optimization model and the flight restricted area information set, wherein the trajectory optimization model comprises a long short-term memory network, a fully connected layer, and a neural ordinary differential equation network; input the predicted flight state feature into the neural ordinary differential equation network to obtain an optimized control parameter; control the flight logistics device according to the optimized control parameter, and determine a predicted flight state vector corresponding to the optimized control parameter based on a preset flight dynamics equation; generate an updated flight state vector sequence according to the predicted flight state vector and the flight state vector sequence; and a second execution unit configured to perform the trajectory optimization steps again with the updated flight state vector sequence as the flight state vector sequence.

[0009] In a third aspect, some embodiments of the present disclosure provide an electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any of the implementations of the first aspect.

[0010] In a fourth aspect, some embodiments of the present disclosure provide a computer readable medium having stored thereon a computer program, wherein the program, when executed by a processor, implements the method described in any implementation manner of the first aspect.

[0011] The above various embodiments of the present disclosure have the following beneficial effects: the method for logistics track optimization of the industry chain according to some embodiments of the present disclosure can reduce flight energy consumption. Specifically, the reason for the increase in flight energy consumption is that numerical methods usually require complex integral operations, rely on artificial experience to design flight corridors, consume a lot of manpower and computing resources, and lack a dynamic adjustment mechanism, making it difficult to achieve global optimal trajectory planning when bypassing the no-fly zone, resulting in prolonged transportation time. Based on this, the method for logistics track optimization of the industry chain according to some embodiments of the present disclosure first detects the no-fly zone of the flight area of the flight logistics equipment to obtain a set of no-fly zone information. Second, the flight state vector sequence of the flight logistics equipment is obtained. Then, in response to the flight remaining range corresponding to the flight logistics equipment being greater than a preset distance value, the following trajectory optimization steps are performed: first, based on the trajectory optimization model and the set of no-fly zone information, the predicted flight state features corresponding to the flight state vector sequence are determined. The trajectory optimization model includes a long short-term memory network, a fully connected layer, and a neural ordinary differential equation network. The trajectory optimization model trained by reinforcement learning is used for prediction, which can dynamically adjust the flight path. Reinforcement learning training enables the model to adjust the flight path according to real-time feedback to ensure global optimal trajectory planning when bypassing the no-fly zone, thereby reducing unnecessary flight time and energy consumption. Then, the predicted flight state features are input into the neural ordinary differential equation network to obtain optimized control parameters. In this way, more accurate optimized control parameters can be obtained, which can effectively adjust the control strategy of the flight equipment to maximize energy efficiency under different flight states. Next, the flight logistics equipment is controlled according to the optimized control parameters, and the predicted flight state vector corresponding to the optimized control parameters is determined based on a preset flight dynamics equation. Then, the updated flight state vector sequence is generated according to the predicted flight state vector and the flight state vector sequence. Finally, the updated flight state vector sequence is used as the flight state vector sequence to perform the trajectory optimization steps again. By dynamically updating the flight state vector sequence and re-executing the optimization steps, the flight path can be adjusted in real time according to changes in the flight environment, avoiding the problem of efficiency decline or energy consumption increase that may occur during flight. BRIEF DESCRIPTION OF DRAWINGS

[0012] The above and other features, advantages, and aspects of embodiments of the present disclosure will become more apparent by describing in detail some embodiments thereof with reference to the annexed drawings in which: like reference numerals refer to like elements throughout. The annexed drawings are schematic and are not intended to accurately depict the proportions or relative positions of elements.

[0013] Figure 1 is a flowchart of some embodiments of a logistics trajectory optimization method for an industrial chain according to the present disclosure;

[0014] Figure 2 is an example diagram of a flyable area of some embodiments of a logistics trajectory optimization method for an industrial chain according to the present disclosure;

[0015] Figure 3 is a flight trajectory diagram of some embodiments of a logistics trajectory optimization method for an industrial chain according to the present disclosure;

[0016] Figure 4 is a structural schematic diagram of some embodiments of a logistics trajectory optimization apparatus for an industrial chain according to the present disclosure;

[0017] Figure 5 is a structural schematic diagram of an electronic device suitable for use to implement some embodiments of the present disclosure. DETAILED DESCRIPTION

[0018] Embodiments of the present disclosure will be described hereinafter with reference to the accompanying drawings. While certain embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and fully convey the scope of the present disclosure to those skilled in the art. It should be understood that the drawings and embodiments are only for illustrative purposes and are not intended to limit the scope of protection of the present disclosure.

[0019] It should also be noted that, for the sake of brevity, only some of the various embodiments of the present disclosure are specifically described herein. It is also understood that the embodiments and features of the present disclosure can be combined with each other, where possible.

[0020] It should be noted that the terms “first”, “second”, and the like in the present disclosure are used only to distinguish different devices, modules, or units, and do not imply the order or interdependence of the functions performed by these devices, modules, or units.

[0021] It should be noted that the terms “one”, “multiple”, and the like in the present disclosure are illustrative and not limiting, and those skilled in the art should understand that, unless otherwise explicitly stated in the context, “one” or “multiple” should be understood as “one or more”.

[0022] Names of messages or information exchanged between a plurality of devices in the embodiments of the present disclosure are only for illustrative purposes, and are not intended to limit the scope of the messages or information.

[0023] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0024] Figure 1 Flow 100 of some embodiments of a logistics trajectory optimization method for an industrial chain according to the present disclosure is shown. The logistics trajectory optimization method for an industrial chain includes the following steps:

[0025] Step 101, detecting a flight restricted area of a flight logistics device to obtain a flight restricted area information set.

[0026] In some embodiments, the execution subject of the logistics trajectory optimization method for an industrial chain can detect a flight restricted area of a flight logistics device to obtain a flight restricted area information set. The execution subject can be a computing device. The flight logistics device can be a flight vehicle for logistics transportation, such as a logistics transportation aircraft.

[0027] In some optional implementations of some embodiments, the execution subject detecting a flight restricted area of a flight logistics device to obtain a flight restricted area information set can include the following steps:

[0028] First, flight condition information corresponding to the flight logistics device is obtained. The flight condition information can include flight start coordinates, flight end coordinates, and aircraft parameter information. The flight start coordinates can be the longitude, latitude, and altitude of the starting point of the current flight task of the flight logistics device. The flight end coordinates can be the longitude, latitude, and altitude of the end point of the flight task. The aircraft parameter information can include the maximum horizontal flight altitude (absolute ceiling) corresponding to the flight logistics device.

[0029] Second, flight region information corresponding to the flight logistics device is generated based on the flight condition information. The flight region information can represent a cylindrical region between the flight start coordinates and the flight end coordinates. In practice, first, the midpoint between the flight start coordinates and the flight end coordinates can be determined as the center of the base circle of the cylindrical region. Second, the distance between the flight start coordinates and the flight end coordinates can be determined as the diameter of the base circle of the cylindrical region. Then, the maximum horizontal flight altitude of the flight logistics device can be determined as the height of the cylindrical region.

[0030] In a third step, according to the flight area information, the preset flight prohibition information set is determined as the flight prohibition area information set, which satisfies the preset condition. Each flight prohibition information in the flight prohibition information set can represent a flight prohibition area. The flight prohibition information can include the centroid coordinates of the flight prohibition area, the radius of the flight prohibition area, and the like. The flight prohibition information set can be obtained from the navigation notice published by the aviation management agency. In practice, the flight prohibition information set that overlaps with the cylindrical area can be determined as the flight prohibition area information, and the flight prohibition area information set is obtained.

[0031] In step 102, a flight state vector sequence of the flight logistics equipment is obtained.

[0032] In some embodiments, the execution subject can obtain the flight state vector sequence of the flight logistics equipment. In practice, first, the flight data of the flight logistics equipment at the same time (including: flight start height, flight start speed, flight start longitude, flight start latitude, flight end longitude, flight end latitude, aircraft geocentric distance, aircraft longitude, aircraft latitude, flight speed, aircraft track angle, and aircraft heading angle) can be obtained, and each flight data is determined as a vector to obtain a flight state vector. Secondly, the flight state vectors in a time period when the flight logistics equipment performs a flight task are determined as a flight state vector sequence. The flight state vector sequence can include 5 flight state vectors, which are not limited here.

[0033] In step 103, in response to the corresponding flight remaining range of the flight logistics equipment being greater than a preset distance value, the following trajectory optimization steps are performed:

[0034] In step 1031, a predicted flight state feature corresponding to the flight state vector sequence is determined based on the trajectory optimization model and the flight prohibition area information set.

[0035] In some embodiments, the execution subject can determine a predicted flight state feature corresponding to the flight state vector sequence based on the trajectory optimization model and the set of no-fly zone information. The flight remaining range can be a distance value between the flight logistics device and the flight end coordinate. The preset distance value can be a length value. For example, the distance value can be 500 m, which is not limited herein. The trajectory optimization model can include a long short-term memory network, a fully connected layer, and a neural ordinary differential equation network. The trajectory optimization model can be a trajectory optimization model trained by reinforcement learning. The long short-term memory network can be a long short-term memory network trained by reinforcement learning, such as an LSTM network. The fully connected layer can be a fully connected layer (FC) trained by reinforcement learning. The neural ordinary differential equation network can be a neural ordinary differential equation network (Neural ODE) trained by reinforcement learning.

[0036] In practice, first, the current longitude, latitude, and altitude of the flight logistics device can be obtained. Then, the distance between the flight logistics device and the flight end coordinate can be determined based on the current longitude, latitude, and altitude of the flight logistics device to obtain the flight remaining range. Next, when the flight remaining range corresponding to the flight logistics device is greater than the preset distance value, the trajectory optimization step is performed.

[0037] In some optional implementations of some embodiments, the execution subject can determine a predicted flight state feature corresponding to the flight state vector sequence based on the trajectory optimization model and the set of no-fly zone information, which can include the following steps:

[0038] First, time series feature vectors are extracted from the flight state vector sequence based on the long short-term memory network and the set of no-fly zone information. In practice, the flight state vector sequence and the set of no-fly zone information can be input into the long short-term memory network trained by reinforcement learning to obtain the set of time series feature vectors. Each flight state feature vector in the flight state vector sequence can correspond to a time series feature vector.

[0039] Second, the time series feature vector corresponding to the last time series feature vector in the set of time series feature vectors is determined as a hidden feature vector.

[0040] Third, the hidden feature vector is mapped to a predicted flight state feature based on the fully connected layer. The predicted flight state feature can represent the flight state at a future time. The predicted flight state feature can be in the form of a vector. In practice, the hidden feature vector can be input into the fully connected layer to obtain the predicted flight state feature.

[0041] Step 1032, input the predicted flight state feature into the neural ordinary differential equation network to obtain the optimized control parameter.

[0042] In some embodiments, the execution body can input the predicted flight state feature into the neural ordinary differential equation network to obtain the optimized control parameter. The optimized control parameter can represent the control parameter of the next time step. The optimized control parameter can include the aircraft roll angle and the aircraft angle of attack.

[0043] Step 1033, control the flight logistics equipment according to the optimized control parameter, and determine the predicted flight state vector corresponding to the optimized control parameter based on the preset flight dynamics equation.

[0044] In some embodiments, the execution body can control the flight logistics equipment according to the optimized control parameter, and determine the predicted flight state vector corresponding to the optimized control parameter based on the preset flight dynamics equation. The predicted flight state vector can represent the flight state of the flight logistics equipment at the next time step. The values in the predicted flight state vector can represent the initial flight altitude, the initial flight speed, the initial flight longitude, the initial flight latitude, the terminal flight longitude, the terminal flight latitude, the aircraft geocentric distance at the next time step, the aircraft longitude at the next time step, the aircraft latitude at the next time step, the flight speed at the next time step, the aircraft track angle at the next time step, and the aircraft heading angle at the next time step. In practice, first, the flight logistics equipment can be controlled according to the optimized control parameter, so that the roll angle and the angle of attack of the flight logistics equipment are equal to the aircraft roll angle and the aircraft angle of attack contained in the optimized control parameter. Here, the flight logistics equipment can include a control component. The control component can adjust the aircraft roll angle and the aircraft angle of attack of the flight logistics equipment according to the optimized control parameter. Then, based on the following flight dynamics equation, the predicted flight state vector corresponding to the optimized control parameter can be determined according to the values in the last flight state vector in the flight state vector sequence:

[0045] .

[0046] wherein, , , , , , respectively represent the aircraft geocentric distance at the last time step, the aircraft longitude at the last time step, the aircraft latitude at the last time step, the flight speed at the last time step, the aircraft track angle at the last time step, and the aircraft heading angle at the last time step. , , , , respectively represent the aircraft geocentric distance, the aircraft longitude, the aircraft latitude, the flight speed, the aircraft track angle, and the aircraft heading angle of the next time step. respectively represent the lift force and the drag force acting on the aircraft, and can be determined according to the above aircraft angle of attack, the flight speed of the previous time step, and the formula in sub-step five in the following optional step. represents the weight of the above flight stream device. represents the aircraft roll angle. represents the time step, which can be 0.1s. represents the reference gravitational acceleration, which can be the surface gravity acceleration of the Earth.

[0047] Step 1034, generating an updated flight state vector sequence according to the predicted flight state vector and the flight state vector sequence.

[0048] In some embodiments, the execution subject can generate an updated flight state vector sequence according to the predicted flight state vector and the flight state vector sequence. In practice, the sequence composed of each flight state vector in the flight state vector sequence except the first flight state vector and the predicted flight state vector can be determined as the updated flight state vector sequence.

[0049] Step 104, executing the trajectory optimization step again with the updated flight state vector sequence as the flight state vector sequence.

[0050] In some embodiments, the execution subject can execute the trajectory optimization step again with the updated flight state vector sequence as the flight state vector sequence.

[0051] Optionally, the trajectory optimization model can be trained by the following steps:

[0052] First, generating a sample training data set based on a numerical method. In practice, the sample training data set can be generated by a numerical method according to the flight dynamics equation of the aircraft. Each sample training data in the sample training data set can include a sample flight vector sequence and a true value vector sequence. Each sample flight vector sequence corresponds to a true value vector sequence. The sample flight vector sequence and the corresponding true value vector sequence can jointly represent the flight state of the aircraft within a period of time.

[0053] ​​Secondly, the initial trajectory optimization model is pre-trained according to the sample training data set to obtain a pre-trained trajectory optimization model. The initial trajectory optimization model includes an initial long short-term memory network, an initial full connection layer, and an initial neural ordinary differential equation network. The initial long short-term memory network can be an LSTM network. The initial full connection layer can be an untrained full connection layer (FC). The initial neural ordinary differential equation network can be a Neural ODE network. In practice, each sample flight vector sequence corresponding to each sample training data in the sample training data set can be taken as an input of the initial trajectory optimization model, and each true value vector sequence corresponding to each sample training data in the sample training data set can be taken as an expected output of the initial trajectory optimization model, so as to train the initial trajectory optimization model to obtain the pre-trained trajectory optimization model.

[0054] Thirdly, the pre-trained trajectory optimization model is trained by reinforcement learning to obtain a trajectory optimization model. In practice, the pre-trained trajectory optimization model can be trained by a reinforcement learning model training method to obtain the trajectory optimization model.

[0055] In practice, when the sample training data is generated, a common technical problem is that a large amount of calculation and a long calculation time are consumed when the bank angle of the aircraft is solved by using an integral algorithm such as a predictor-corrector algorithm or a pseudospectral method, resulting in waste of calculation resources.

[0056] Optionally, the execution subject generates the sample training data set based on the numerical method, and can include the following steps:

[0057] Firstly, an aircraft trajectory data set is obtained. Each aircraft trajectory data in the aircraft trajectory data set can include a sample aircraft state quantity sequence. The elements in the sample aircraft state quantity sequence can be arranged in chronological order. Each sample aircraft state quantity in the sample aircraft state quantity sequence can include a sample flight starting height, a sample flight starting speed, a sample flight starting longitude, a sample flight starting latitude, a sample flight terminal longitude, a sample flight terminal latitude, a sample aircraft geocentric distance, a sample aircraft longitude, a sample aircraft latitude, a sample flight speed, a sample aircraft track angle, and a sample aircraft heading angle.

[0058] Secondly, for each aircraft trajectory data in the aircraft trajectory data set, the following steps are performed to generate sample training data, and a sample training data set is obtained:

[0059] Firstly, for each sample aircraft state quantity in the sample aircraft state quantity sequence included in the aircraft trajectory data, the following steps are performed:

[0060] Sub-step one, the sample aircraft geocentric distance and the sample flight speed contained in the sample aircraft state quantity are dimensionless processed to obtain a dimensionless height and a dimensionless speed. In practice, the sample aircraft geocentric distance and the sample flight speed contained in the sample aircraft state quantity can be dimensionless processed by the following formula to obtain the dimensionless height and the dimensionless speed.

[0061] .

[0062] wherein z represents the dimensionless height. r represents the sample aircraft geocentric distance. represents the reference length in the dimensionless processing, and the reference length can be the earth radius. u represents the dimensionless speed. V represents the sample flight speed. represents the reference speed in the dimensionless processing. represents the reference gravitational acceleration. represents the dimensionless time, and t represents the length of a time step in the flight of the aircraft, for example, 0.1s.

[0063] Sub-step two, the sample aircraft energy corresponding to the dimensionless height and the dimensionless speed is determined. In practice, the sample aircraft energy corresponding to the dimensionless height and the dimensionless speed can be determined by the following formula.

[0064] .

[0065] wherein, represents the sample aircraft energy. When the z is the sample flight terminal height and the u is the sample flight terminal speed, the result of the formula can represent the sample aircraft terminal energy . Here, the sample aircraft state quantity corresponds to a flight terminal. The flight terminal is the endpoint of the current flight task of the sample aircraft. The flight terminal can correspond to a sample flight terminal longitude, a sample flight terminal latitude, a sample flight terminal height, and a sample flight terminal speed. The sample flight terminal height can be the height of the endpoint of the current flight task of the sample aircraft. The sample flight terminal speed can be a preset speed corresponding to the endpoint of the current flight task.

[0066] Sub-step three, the atmospheric density in which the aircraft is located is generated according to the sample aircraft geocentric distance corresponding to the sample aircraft state quantity. In practice, the difference between the sample aircraft geocentric distance and the earth radius can be determined as the sample aircraft height. Then, the atmospheric density in which the aircraft is located corresponding to the sample aircraft state quantity can be determined according to a preset height-atmospheric density mapping table.

[0067] Sub-step four, according to the preset maximum angle of attack, the preset maximum lift-drag ratio angle of attack, the preset maximum angle of attack speed value, the preset maximum lift-drag ratio angle of attack speed value, the preset terminal speed, the sample aircraft state quantity includes the sample aircraft angle of attack corresponding to the sample flight speed. Wherein, the sample aircraft angle of attack can represent the angle between the incoming flow and the wing chord, when the wing is upward, it is positive angle of attack, when it is downward, it is negative angle of attack. In practice, the sample aircraft angle of attack corresponding to the sample aircraft state quantity can be determined by the following formula.

[0068] .

[0069] Wherein, represents the sample aircraft angle of attack. represents the preset maximum angle of attack. represents the maximum lift-drag ratio angle of attack. represents the maximum angle of attack speed lower limit, which can be the speed value preset under the maximum angle of attack. represents the maximum lift-drag ratio angle of attack speed upper limit, which can be the speed value preset under the maximum lift-drag ratio angle of attack. represents the sample flight terminal speed.

[0070] In practice, the maximum angle of attack can be 20 degrees. The maximum angle of attack speed lower limit can be 5000 meters per second. The maximum lift-drag ratio angle of attack can be 12 degrees. The maximum lift-drag ratio angle of attack speed upper limit can be 2500 meters per second.

[0071] Sub-step five, according to the atmospheric density where the aircraft is located, the sample aircraft angle of attack, the sample flight speed contained in the sample aircraft state quantity, and the preset aircraft characteristic area, the lift acting on the aircraft and the drag acting on the aircraft are generated. In practice, the lift acting on the aircraft and the drag acting on the aircraft can be generated by the following formula:

[0072] .

[0073] Wherein, L represents the lift acting on the aircraft. D represents the drag acting on the aircraft. represents the lift coefficient. represents the drag coefficient. represents the atmospheric density where the aircraft is located. all represent coefficients, which are not specifically limited here. e represents the natural constant. S represents the aircraft characteristic area, which represents the sample aircraft wind area (m 2 ).

[0074] Sub-step six, according to the sample flight terminal longitude and the sample flight terminal latitude included in the sample aircraft state quantity, determine the distance to be flown corresponding to the sample aircraft state quantity. Wherein, the distance to be flown can be a circular arc formed by the current aircraft longitude and latitude and the terminal longitude and latitude. In practice, the distance to be flown corresponding to the sample aircraft state quantity can be determined by the following formula:

[0075] .

[0076] Wherein, indicates the distance to be flown corresponding to the sample aircraft state quantity. indicates the sample aircraft longitude. indicates the sample flight terminal longitude. indicates the sample aircraft latitude. indicates the sample flight terminal latitude.

[0077] Sub-step seven, according to the distance to be flown, the lift force of the aircraft, the drag force of the aircraft, the sample aircraft energy, and the dimensionless height, generate the sample aircraft roll angle corresponding to the sample aircraft state quantity. Wherein, the sample aircraft roll angle can represent the angle of rotation around the longitudinal axis during the flight of the aircraft. In practice, first, the sample aircraft roll angle value corresponding to the sample aircraft state quantity can be generated by the following formula.

[0078] .

[0079] Wherein, indicates the sample aircraft roll angle.

[0080] Secondly, the azimuth angle of the sample aircraft can be determined by the following formula:

[0081] .

[0082] Wherein, indicates the azimuth angle.

[0083] Then, when the difference between the azimuth angle and the heading angle corresponding to the sample aircraft is equal to the preset angle value, it indicates that the sample aircraft triggers the lateral motion boundary, and the sign of the sample aircraft roll angle can be flipped, that is, the sign of the sample aircraft roll angle at this time step can be opposite to the sign of the sample aircraft roll angle at the last time step. Here, the preset angle value can be a preset degree, which is not limited here. The sample aircraft roll angle at the first time step can be determined by the prediction correction algorithm and the pseudo-spectral method integral algorithm.

[0084] In practice, the flight process is in the energy descent process, and the value of the tilt angle is solved by energy and the distance to be flown, so that the remaining range is equal to the distance to be flown. The above-mentioned remaining range can represent the actual distance that the sample aircraft can fly at the current energy.

[0085] Sub-step eight, the above-mentioned sample aircraft attack angle and the above-mentioned sample aircraft tilt angle are determined as the sample aircraft control quantity corresponding to the sample aircraft state quantity. Wherein, the above-mentioned sample aircraft control quantity can include sample aircraft tilt angle and sample aircraft attack angle.

[0086] In practice, the friction between the wing and the airflow during the gliding process of the aircraft generates aerodynamic heat, and in order to ensure the overall safety of the fuselage, it is necessary to limit the maximum heat flux rate. In addition, the lift, drag and gravity act on the aircraft at the same time, preventing the resultant force from exceeding the upper limit of the material strength, and it is necessary to limit the maximum overload and dynamic pressure. The above-mentioned dimensionless height and dimensionless speed need to meet the following constraint conditions:

[0087] .

[0088] Wherein, The heat flux rate is represented by. The maximum limited heat flux rate (kw / m 2 ). The dynamic pressure is represented by. The maximum limited dynamic pressure (kPa) is represented by. The overload is represented by. The maximum limited overload is represented by.

[0089] Second sub-step, the above-mentioned sample aircraft state quantity sequence and the sample aircraft control quantity sequence corresponding to each sample aircraft control quantity determined are combined into sample training data.

[0090] The first step to the second step and the related content thereof are an application point of one embodiment of the present disclosure, and solve the technical problem of waste of computing resources. Factors that cause the technical problem are usually as follows: when an integral algorithm such as a predictor-corrector algorithm or a pseudospectral method is used to solve the numerical value of the roll angle of the aircraft, a large amount of computation and a long computation time are often consumed. If the factors are solved, the flight time and the consumption of energy during flight can be reduced. To achieve this effect, first, for each sample aircraft state quantity in the sample aircraft state quantity sequence, the altitude and the speed contained in the sample aircraft state quantity can be dimensionless processed to reduce the workload. Then, the sample aircraft angle of attack, the lift acting on the aircraft, and the drag acting on the aircraft can be determined according to a preset formula. After that, the energy of the sample aircraft at the current position can be determined according to the dimensionless altitude and the dimensionless speed corresponding to the current sample aircraft. In this way, the predicted remaining range can be analyzed based on energy. Then, the circular arc between the longitude and latitude included in the current sample aircraft state quantity and the terminal longitude and latitude is determined as the distance to be flown. Secondly, during flight, because it is in the energy consumption stage, the optimal roll angle that matches the remaining range and the distance to be flown can be derived based on the constraint relationship between the current remaining energy and the distance to be flown, in combination with the lift acting on the aircraft and the drag acting on the aircraft. Compared with the integral algorithms such as the predictor-corrector algorithm and the pseudospectral method, the numerical value of the roll angle solved by the energy method can greatly reduce the computation amount and the computation time, thereby reducing the waste of computing resources.

[0091] Optionally, the execution subject pre-trains the preset initial trajectory optimization model according to the sample training data set to obtain a pre-trained trajectory optimization model, which can include the following steps:

[0092] First, each sample training data in the sample training data set is preprocessed to generate preprocessed sample training data, and a preprocessed sample training data set is obtained. Each preprocessed sample training data in the preprocessed sample training data set includes a preprocessed sample state quantity sequence and a preprocessed control quantity true value. In practice, first, for each sample training data in the sample training data set, a sequence composed of a plurality of continuous sample aircraft state quantities in the sample aircraft state quantity sequence corresponding to the sample training data can be determined as a preprocessed sample state quantity sequence, and a sample aircraft control quantity at the next time step corresponding to the plurality of continuous sample aircraft state quantities can be determined as a preprocessed control quantity true value.

[0093] In the second step, the preprocessed sample state sequence included in each preprocessed sample training data in the preprocessed sample training data set is input into a preset initial trajectory optimization model to obtain a predicted control quantity set. Each predicted control quantity in the predicted control quantity set can represent a flight vehicle control parameter at a future time step. The predicted control quantity can represent a roll angle and an attack angle of the flight vehicle. In practice, each preprocessed sample training data in the preprocessed sample training data set can be input into an initial long short-term memory network and an initial fully connected layer included in the initial trajectory optimization model to obtain a sample predicted feature set. Then, each sample predicted feature in the sample predicted feature set is input into an initial neural ordinary differential equation network included in the initial trajectory optimization model to obtain a sample control parameter sequence set.

[0094] In practice, the initial neural ordinary differential equation network can include a preset fitting algorithm. The fitting algorithm can be an Euler algorithm or a Runge-Kutta algorithm. The fitting algorithm can be used to approximate a primitive function of a roll angle of the flight vehicle with respect to time by an iterative method. The primitive function of the roll angle of the flight vehicle with respect to time is shown in the following formula:

[0095] .

[0096] wherein, represents a primitive function of a roll angle of the flight vehicle with respect to time t. represents a differential of the roll angle. represents a derivative of the roll angle with respect to time, which is represented by a neural network. The neural network can be a fully connected network. p represents a trainable parameter of the neural network.

[0097] In practice, after the primitive function is calculated by the fitting algorithm, a roll angle value at a desired time point can be arbitrarily specified to achieve continuous prediction.

[0098] In the third step, a total loss value is generated according to a loss value between each predicted control quantity in the predicted control quantity set and a preprocessed control quantity true value included in corresponding preprocessed sample training data in the preprocessed sample training data set. In practice, first, a loss value between each predicted control quantity in the predicted control quantity set and a preprocessed control quantity true value included in corresponding preprocessed sample training data in the preprocessed sample training data set can be determined by a preset loss algorithm. The loss algorithm can be a mean square error (MSE) algorithm. Then, a sum of the determined loss values is determined as the total loss value.

[0099] In the fourth step, a pre-training trajectory optimization model is obtained by performing back propagation parameter update on the initial trajectory optimization model according to the total loss value.

[0100] In practice, when encountering a no-fly area, the trajectory optimization method based on numerical operation often needs to solve a highly complex and mutually coupled system of differential equations of dynamic systems, which is time-consuming and difficult to effectively avoid the no-fly area while shortening the detour path as much as possible, resulting in a longer flight route of the flight device, thereby increasing the flight time and energy consumption during flight.

[0101] Optionally, the execution subject can perform reinforcement learning training on the pre-trained trajectory optimization model to obtain a trajectory optimization model, which can include the following steps: First, determine the sample flight start point and sample flight end point corresponding to the sample aircraft. The sample aircraft can be a flight simulation device in a simulation environment. The sample flight start point can be the starting point of the flight task performed by the sample aircraft, which can be represented by longitude, latitude, and altitude. The sample flight end point can be the end point of the flight task performed by the sample aircraft, which can be represented by longitude, latitude, and altitude.

[0103] Second, generate a sample no-fly area set according to the sample flight start point and the sample flight end point. The sample no-fly area set can include at least one sample no-fly area, which can be represented by the longitude and latitude of the centroid and the radius. In practice, at least one coordinate point can be randomly generated within the boundary region between the sample flight start point and the sample flight end point. Each coordinate point in the at least one coordinate point corresponds to a radius. Then, according to each coordinate point in the at least one coordinate point and its corresponding radius, a sample no-fly area can be determined to obtain a sample no-fly area set. Here, the boundary region between the sample flight start point and the sample flight end point can be a circular region in a two-dimensional bird's eye view with the line segment between the sample flight start point and the sample flight end point as the diameter.

[0104] Third, based on a predetermined number of rounds, perform the following repeated training steps:

[0105] A first sub-step, generating a sample flight state vector sequence according to the sample flight start point and the sample flight end point. The round can refer to the number of pre-set repeated training. The pre-set round can be 200 rounds, which is not limited here. The sample flight state vector sequence can be a sequence composed of a plurality of continuous time step sample flight state vectors. The values in each sample flight state vector in the sample flight state vector sequence can represent the sample flight start point height, the sample flight start point speed, the sample flight start point longitude, the sample flight start point latitude, the sample flight end point longitude, the sample flight end point latitude, the sample flight vehicle geocentric distance, the sample flight vehicle longitude, the sample flight vehicle latitude, the sample flight speed, the sample flight vehicle track angle and the sample flight vehicle heading angle. In practice, first, the starting sample flight state vector can be determined. Specifically, the starting sample flight state vector can be determined according to the longitude, latitude and height corresponding to the sample flight start point, the longitude, latitude and height corresponding to the sample flight end point. Secondly, the sample flight vehicle geocentric distance, the sample flight vehicle longitude and the sample flight vehicle latitude corresponding to the starting sample flight state vector can be the same as the values of the sample flight start point height, the sample flight start point longitude and the sample flight start point latitude. Then, the angle between the line segment between the sample flight start point and the sample flight end point and the line segment in the north direction on the ground plane can be determined as the sample flight vehicle track angle corresponding to the starting sample flight state vector. The value of the sample flight vehicle heading angle corresponding to the starting sample flight state vector can be the same as the value of the sample flight vehicle track angle. The sample flight start point speed corresponding to the starting sample flight state vector can be a pre-set speed value. The value of the sample flight speed corresponding to the starting sample flight state vector can be the same as the value of the sample flight start point speed. Then, the sample flight state vector sequence can be generated according to the starting sample flight state vector. Specifically, a plurality of sample flight state vectors can be obtained by copying the starting sample flight state vector to generate the sample flight state vector sequence.

[0106] A second sub-step, in response to determining that the sample flight state vector corresponding to the last one in the sample flight state vector sequence does not satisfy the pre-set end condition, performing the following reinforcement training steps based on the pre-trained trajectory optimization model:

[0107] Sub-step one, determining a sample predicted control amount corresponding to the sample flight state vector sequence according to a pre-trained trajectory optimization model. The end condition can include a no-fly constraint condition, a boundary constraint condition, a flyable region constraint condition, and a normal end condition. The no-fly constraint condition can be that the longitude and latitude corresponding to the sample aircraft of the sample flight state vector are not in each sample no-fly region in the sample no-fly region set, which can be expressed by the following formula:

[0108] .

[0109] wherein, represents the latitude of the centroid of the no-fly region. represents the longitude of the centroid of the no-fly region. represents the radius of the no-fly region.

[0110] The boundary constraint condition can be that the longitude and latitude corresponding to the sample aircraft cannot be outside the boundary region. The flyable region constraint condition can be that the longitude and latitude corresponding to the sample aircraft cannot be outside the preset flyable region. The flyable region can be the region between the line segment between the sample flight starting point and the sample flight ending point and the preset parallel line. Specifically, the line segment between the sample flight starting point and the sample flight ending point can be determined as the target line segment. The preset parallel line can be a line segment parallel to the target line segment and having a distance from each sample no-fly region in the sample no-fly region set higher than a preset distance threshold. The distance threshold can be 1 km, which is not limited here. An example diagram corresponding to the flyable region is shown in Figure 2 The normal end condition can be that the distance between the sample aircraft and the sample flight ending point is less than a preset end distance threshold. The end distance threshold can be 0.5 km, which is not limited here.

[0111] In practice, first, for each sample flight state vector in the sample flight state vector sequence, the sample flight state vector and each no-fly region in the sample no-fly region set can be combined to form a state space vector to generate a state space vector sequence. Each state space vector in the state space vector sequence can include a sample flight state vector, a centroid coordinate of each no-fly region, and a radius. Then, the state space vector sequence can be input into the pre-trained trajectory optimization model to obtain a sample predicted control amount. The sample predicted control amount can include a predicted bank angle and a predicted attack angle. Here, the end condition can also include a bank angle constraint condition. The bank angle constraint condition can limit the activity range of the bank angle to between -50 degrees and 50 degrees.

[0112] Sub-step two, based on the flight dynamics equation, determine the flight state vector corresponding to the next time of the sample prediction control amount. In practice, the flight state vector corresponding to the next time of the sample prediction control amount can be determined based on the flight dynamics equation through the content in step 1033. Details are not repeated here.

[0113] Sub-step three, determine the reward value corresponding to the next time flight state vector, and update the parameters of the pre-trained trajectory optimization model according to the reward value to obtain an updated trajectory optimization model. The reward value can be the sum of distance reward, end reward and time reward. The distance reward can be represented by the following formula:

[0114] .

[0115] Wherein, The distance reward is represented by the following formula: The distance to be flown is represented by the following formula: The overall flight distance is represented by the following formula:

[0116] The end reward can be determined as a preset end reward value when the distance between the sample aircraft and the sample flight endpoint is less than the end distance threshold. The end reward value can be a positive number, which is not specifically limited here.

[0117] The time reward can be represented by the following formula:

[0118] .

[0119] Wherein, The time reward is represented by the following formula: The time reward coefficient is represented by the following formula: The time step corresponding to the next time flight state vector is represented by the following formula: The maximum flight time step is represented by the following formula: Here, the time reward coefficient can be a natural number, and its specific value can be determined by cross-validation experiment, which is not specifically limited here. Here, the time reward can avoid the model in the reinforcement learning training to consume time meaninglessly for the purpose of obtaining distance reward, so as to reduce the training time.

[0120] In practice, firstly, the distance to be flown between the sample aircraft and the sample flight destination can be generated based on the sample aircraft's geocentric distance, longitude, and latitude corresponding to the next-moment flight state vector, and the longitude, latitude, and altitude corresponding to the sample flight destination. Then, the distance reward can be obtained using the formula for generating the distance reward. Secondly, a time reward can be generated based on the formula for generating the time reward and the time step corresponding to the next-moment flight state vector. Next, when the distance to be flown is less than the termination distance threshold, the termination reward value is determined as the termination reward; otherwise, the termination reward is 0. Then, the sum of the distance reward, the time reward, and the termination reward can be determined as the reward value. Finally, the parameters of the pre-trained trajectory optimization model can be updated using a preset parameter update technique to obtain an updated trajectory optimization model. Here, the parameter update technique can be the value function method (Q-learning) or the policy gradient method (REINFORCE).

[0121] Sub-step four: Based on the aforementioned next-moment flight state vector and the aforementioned sample flight state vector sequence, generate a second flight state vector sequence. In practice, the second flight state vector sequence can be determined by the sequence formed by each sample flight state vector (excluding the first sample flight state vector) in the aforementioned sample flight state vector sequence and the aforementioned next-moment flight state vector.

[0122] Sub-step five: In response to the determination that the flight state vector at the next moment does not satisfy the termination condition, the updated trajectory optimization model is used as a pre-trained trajectory optimization model, and the second flight state vector sequence is used as a sample flight state vector sequence, and the reinforcement training step is executed again. In practice, firstly, the geocentric distance, longitude, and latitude of the sample aircraft included in the flight state vector at the next moment can be used to determine whether the flight state vector at the next moment satisfies the termination condition. Secondly, when the flight state vector at the next moment does not satisfy the termination condition, the updated trajectory optimization model can be used as a pre-trained trajectory optimization model, and the second flight state vector sequence can be used as a sample flight state vector sequence, and the reinforcement training step is executed again.

[0123] The third sub-step, in response to determining that the flight state vector at the next moment satisfies the termination condition and the current round is not equal to the preset round, terminates the reinforcement training step and re-executes the repetitive training step. In practice, when the flight state vector at the next moment satisfies the termination condition and the current training round is not equal to the preset round, firstly, the round corresponding to the current reinforcement training step can be terminated. Then, the repetitive training step (i.e., a new round of training) can be executed again.

[0124] Fourthly, the obtained updated trajectory optimization model is determined as the trajectory optimization model. In practice, the updated trajectory optimization model obtained in the last round of training can be determined as the trajectory optimization model.

[0125] Optionally, the execution subject can further perform the following steps:

[0126] Firstly, flight endpoint information, no-fly zone information and at least one flight starting point information are obtained. Each flight starting point information in the at least one flight starting point information can include the starting longitude, starting latitude and starting height of the flight starting point. Each flight starting point information in the at least one flight starting point information can be a coordinate point randomly sampled. The flight endpoint information can include the endpoint longitude, endpoint latitude and endpoint height of the flight endpoint. The no-fly zone information can include the centroid longitude, centroid latitude and no-fly radius of the no-fly zone.

[0127] Secondly, for each flight starting point information in the at least one flight starting point information, the following steps are performed:

[0128] Based on the trajectory optimization model, a sequence of trajectory points is generated according to the flight starting point information, the flight endpoint information and the no-fly zone information. In practice, firstly, a sequence of flight state vectors can be generated according to the flight starting point information and the flight endpoint information in the manner of the first sub-step in the optional step. Details are not repeated here. Secondly, the trajectory optimization step can be performed based on the trajectory optimization model and the no-fly zone information according to the operation in the step 103. For each predicted flight state vector determined in the trajectory optimization step, a trajectory point can be generated according to the corresponding aircraft longitude and aircraft latitude in the predicted flight state vector. Finally, a sequence composed of the generated trajectory points can be determined as the sequence of trajectory points.

[0129] Thirdly, each trajectory point sequence in the generated at least one trajectory point sequence is rendered according to a preset flight map to obtain a flight trajectory map. The flight map can be an overhead view covering the flight endpoint and each flight starting point in the at least one flight starting point. In practice, for each trajectory point sequence in the generated at least one trajectory point sequence, trajectory point plotting can be performed on the flight map according to the longitude and latitude of each trajectory point in the trajectory point sequence to obtain a flight trajectory map. The flight trajectory map is shown in the (a) subgraph of FIG. 1. Figure 3 Figure 3 ​In the above embodiment, (b) each trajectory in the subgraph is obtained by a numerical algorithm. It can be seen that, compared with the flight trajectory obtained by the numerical algorithm, the flight trajectory obtained by the trajectory optimization model has a shorter path, thereby reducing the energy consumption in the flight process.

[0130] The above optional step and its related content serve as one of the invention points of the embodiment of the present disclosure, and solve the above technical problem. The factors that cause the above technical problem are often as follows: when encountering a no-fly area, the trajectory optimization method based on numerical operation often needs to solve a highly complex and mutually coupled system of differential equations, which is time-consuming and difficult to effectively avoid the no-fly area while shortening the detour path as much as possible, resulting in a longer flight route of the flight device. If the above factors are solved, the flight time and energy consumption in the flight process can be reduced. In order to achieve this effect, first, the sample flight starting point and sample flight ending point can be randomly determined, and the no-fly area can be randomly generated. In this way, the effectiveness of the reinforcement learning training can be improved. After generating the no-fly area coordinate information, the trajectory optimization model can generate the optimal trajectory under the condition of not violating the air constraints and reduce the flight time through multiple rounds of reinforcement training according to the preset reward value and end condition. In this way, the trajectory optimization model can generate the optimal trajectory under the condition of not violating the air constraints and reduce the flight time. The reinforcement learning training adopts the pre-trained trajectory optimization model, uses the architecture combining the long short-term memory network and the neural ordinary differential equation, can more effectively process the continuous change of the time series data in the flight process, and make real-time decisions, so that the generated trajectory can better adapt to the environment and effectively avoid the no-fly area. Finally, the flight trajectory can be obtained by drawing the flight trajectory determined by the trajectory optimization model. According to the flight trajectory obtained from different flight starting points, the trajectory optimization model can effectively avoid the no-fly area, and the generated path is shorter, effectively reducing the flight time and energy consumption in the flight process.

[0131] The above various embodiments of the present disclosure have the following beneficial effects: by the logistics track optimization method for the industrial chain of some embodiments of the present disclosure, the flight energy consumption can be reduced. Specifically, the reason for the increase in flight energy consumption is that numerical methods usually require complex integral operations, and rely on artificial experience to design flight corridors, which consumes more manpower and computing resources, and due to the lack of dynamic adjustment mechanism, it is difficult to achieve global optimal trajectory planning when bypassing the no-fly zone, resulting in prolonged transportation time. Based on this, the logistics track optimization method for the industrial chain of some embodiments of the present disclosure first detects the no-fly zone of the flight area of the flight logistics equipment to obtain a set of no-fly zone information. Second, the flight state vector sequence of the flight logistics equipment is obtained. Then, in response to the flight remaining range corresponding to the flight logistics equipment being greater than a preset distance value, the following trajectory optimization steps are performed: first, based on the trajectory optimization model and the set of no-fly zone information, the predicted flight state characteristics corresponding to the flight state vector sequence are determined. The trajectory optimization model includes a long short-term memory network, a fully connected layer, and a neural ordinary differential equation network. The trajectory optimization model trained by reinforcement learning can dynamically adjust the flight path. Reinforcement learning training enables the model to adjust the flight path according to real-time feedback to ensure global optimal trajectory planning when bypassing the no-fly zone, thereby reducing unnecessary flight time and energy consumption. Then, the predicted flight state characteristics are input into the neural ordinary differential equation network to obtain optimized control parameters. In this way, more accurate optimized control parameters can be obtained, which can effectively adjust the control strategy of the flight equipment to maximize energy efficiency under different flight states. Next, the flight logistics equipment is controlled according to the optimized control parameters, and the predicted flight state vector corresponding to the optimized control parameters is determined based on the preset flight dynamics equation. Finally, the updated flight state vector sequence is generated according to the predicted flight state vector and the flight state vector sequence, and the updated flight state vector sequence is used as the flight state vector sequence to perform the trajectory optimization steps again. By dynamically updating the flight state vector sequence and re-executing the optimization steps, the flight path can be adjusted in real time according to the changes in the flight environment, avoiding the problem of efficiency decline or energy consumption increase that may occur during flight.

[0132] Further reference Figure 4 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a logistics track optimization device for the industrial chain, which device embodiments correspond to those method embodiments shown in Figure 1 The device can be applied in various electronic devices.

[0133] As Figure 4As shown, the logistics track optimization apparatus 400 for the industrial chain of some embodiments includes: a no-fly zone detection unit 401, an acquisition unit 402, and an execution unit 403. The no-fly zone detection unit 401 is configured to perform no-fly zone detection on a flight area of a flight logistics device to obtain a set of no-fly zone information. The acquisition unit 402 is configured to acquire a flight state vector sequence of the flight logistics device. The execution unit 403 is configured to, in response to a corresponding flight remaining range of the flight logistics device being greater than a preset distance value, perform the following track optimization steps: determining a predicted flight state feature corresponding to the flight state vector sequence based on a track optimization model and the set of no-fly zone information, wherein the track optimization model includes a long short-term memory network, a fully connected layer, and a neural ordinary differential equation network; inputting the predicted flight state feature into the neural ordinary differential equation network to obtain an optimized control parameter; controlling the flight logistics device according to the optimized control parameter, and determining a predicted flight state vector corresponding to the optimized control parameter based on a preset flight dynamics equation; generating an updated flight state vector sequence according to the predicted flight state vector and the flight state vector sequence; and the second execution unit 404 is configured to perform the track optimization steps again with the updated flight state vector sequence as the flight state vector sequence.

[0134] It can be understood that the units described in the apparatus 400 correspond to the respective steps in the method described above. Figure 1 Therefore, the operations, features, and advantages described above for the method also apply to the apparatus 400 and the units included therein, which will not be described here again.

[0135] Reference is made below to Figure 5 which shows a structural schematic diagram of an electronic device (e.g., a computing device) 500 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not impose any limitation on the functions and use range of the embodiments of the present disclosure.

[0136] As shown in Figure 5 , the electronic device 500 can include a processing apparatus (e.g., a central processor, a graphics processor, etc.) 501, which can perform various appropriate actions and processes according to programs stored in a read-only memory 502 or loaded from a storage device 508 into a random access memory 503. Various programs and data required for the operation of the electronic device 500 are also stored in the random access memory 503. The processing apparatus 501, the read-only memory 502, and the random access memory 503 are connected to each other through a bus 504. An input / output interface 505 is also connected to the bus 504.

[0137] Generally, the following devices can be connected to the I / O interface 505: input device(s) 506, including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, and the like; output device(s) 507, including, for example, a liquid crystal display (LCD), a speaker, a vibrator, and the like; storage device(s) 508, including, for example, a magnetic tape, a hard disk, and the like; and communication device(s) 509. The communication device(s) 509 can allow the electronic device 500 to communicate wirelessly or through a wire with other devices to exchange data. Although Figure 5 The electronic device 500 is shown with various devices, but it is understood that all of the shown devices are not required to be implemented or present. More or fewer devices can alternatively be implemented or present. Figure 5 Each block shown in the flowcharts can represent a device or multiple devices, as desired.

[0138] In particular, processes described above with reference to the flowcharts can be implemented as a computer software program according to some embodiments of the present disclosure. For example, some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program code for executing the methods shown in the flowcharts. In some such embodiments, the computer program can be downloaded and installed from a network through the communication device 509, or installed from the storage device 508, or installed from the read only memory 502. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the methods of some embodiments of the present disclosure are performed.

[0139] Note that the computer readable medium in some embodiments of the present disclosure can be a computer readable signal medium or a computer readable storage medium or any combination thereof. The computer readable storage medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any suitable combination of the foregoing. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In some embodiments of the present disclosure, the computer readable storage medium can be any tangible medium that contains or stores a program used by an instruction execution system, apparatus or device, or that can be used by or in connection with an instruction execution system, apparatus or device. In some embodiments of the present disclosure, the computer readable signal medium can include a computer readable program code that is carried in a baseband or as a part of a carrier wave, in which the computer readable program code can be loaded into a computer readable storage medium. Such a propagated computer readable signal medium can take many forms, including but not limited to, an electromagnetic signal, an optical pulse, or any suitable combination of the foregoing. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport program code for use by or in connection with an instruction execution system, apparatus or device. Program code embodied on a computer readable medium can be transmitted using any suitable medium, including but not limited to, wire, cable, RF (radio frequency), etc., or any suitable combination of the foregoing.

[0140] In some embodiments, the client, server, or both can communicate using any current known or future developed network protocol, such as HTTP (Hyper Text Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include local area networks ("LAN"), wide area networks ("WAN"), the Internet, and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any current known or future developed networks.

[0141] The computer readable medium can be included in the electronic device, or exist separately from the electronic device. The computer readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: perform no-fly zone detection on a flight area of a flight logistics device to obtain a set of no-fly zone information; obtain a sequence of flight state vectors of the flight logistics device; in response to a remaining flight distance corresponding to the flight logistics device being greater than a preset distance value, perform the following trajectory optimization steps: determine predicted flight state features corresponding to the sequence of flight state vectors based on a trajectory optimization model and the set of no-fly zone information, wherein the trajectory optimization model includes a long short-term memory network, a fully connected layer, and a neural ordinary differential equation network; input the predicted flight state features into the neural ordinary differential equation network to obtain optimized control parameters; control the flight logistics device according to the optimized control parameters, and determine predicted flight state vectors corresponding to the optimized control parameters based on a preset flight dynamics equation; generate an updated sequence of flight state vectors based on the predicted flight state vectors and the sequence of flight state vectors; and perform the trajectory optimization steps again using the updated sequence of flight state vectors as the sequence of flight state vectors.

[0142] Computer program code for carrying out operations of some embodiments of the disclosure can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0143] The flow and block diagrams in the drawings represent possible architectural, functional, and operational scenarios of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block can represent a module, a segment, or a portion of code that comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations thereof, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or combinations of hardware and software.

[0144] The units described in some embodiments of the present disclosure can be implemented by means of software, or can be implemented by hardware. The described units can also be arranged in a processor, for example, a processor can be described as including a flight restricted area detection unit, an acquisition unit, an execution unit, and a second execution unit. In some cases, the names of these units do not constitute a limitation on the units themselves, for example, the acquisition unit can also be described as "a unit for acquiring a flight state vector sequence of the flight equipment".

[0145] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, and without limitation, example types of hardware logic components that can be used include Field-programmable Gate Arrays (FPGAs), Application-specific Integrated Circuits (ASICs), Application-specific Standard Products (ASSPs), System-on-a-chip systems (SOCs), Complex Programmable Logic Devices (CPLDs), etc.

[0146] The above description is merely some of the preferred embodiments of the present disclosure and a description of the principles of the technology used. Those skilled in the art should understand that the scope of the application involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or equivalent features without departing from the above inventive concept. For example, the above features can be replaced with technical features disclosed in the embodiments of the present disclosure (but not limited to) having similar functions to form technical solutions.

Claims

1. A method for optimizing logistics trajectories in a supply chain, comprising: No-fly zone detection is performed on the flight area of ​​the flying logistics equipment to obtain a set of no-fly zone information; Obtain the flight state vector sequence of the flying logistics equipment; A sample training dataset is generated based on numerical methods. Each sample training data in the sample training dataset is preprocessed to generate preprocessed sample training data, resulting in a preprocessed sample training dataset. Each preprocessed sample training data in the preprocessed sample training dataset includes a preprocessed sample state sequence and a preprocessed control quantity true value. The preprocessed sample state sequence included in each preprocessed sample training data in the preprocessed sample training dataset is input into a preset initial trajectory optimization model to obtain a set of predictive control variables. The initial trajectory optimization model includes an initial long short-term memory network, an initial fully connected layer, and an initial neural network. The total loss value is generated based on the loss value between each predicted control variable in the predicted control variable set and the true value of the preprocessed control variable included in the corresponding preprocessed sample training data in the preprocessed sample training dataset. The initial trajectory optimization model is backpropagated and updated based on the total loss value to obtain a pre-trained trajectory optimization model. The pre-trained trajectory optimization model is trained using reinforcement learning to obtain the trajectory optimization model; In response to the remaining flight range corresponding to the flying logistics equipment being greater than a preset distance value, the following trajectory optimization steps are performed: Based on the trajectory optimization model and the no-fly zone information set, the predicted flight state features corresponding to the flight state vector sequence are determined. The trajectory optimization model includes a long short-term memory network, a fully connected layer, and a neural network of ordinary differential equations. The predicted flight state features are input into the neural network of constant differential equations to obtain optimized control parameters. The flight logistics equipment is controlled according to the optimized control parameters, and the predicted flight state vector corresponding to the optimized control parameters is determined based on the preset flight dynamics equations. Based on the predicted flight state vector and the flight state vector sequence, an updated flight state vector sequence is generated; The updated flight state vector sequence is then used as the flight state vector sequence to perform the trajectory optimization step again.

2. The method according to claim 1, wherein, The process of detecting no-fly zones in the flight area of ​​the flying logistics equipment to obtain a no-fly zone information set includes: Obtain the flight status information corresponding to the flying logistics equipment, wherein the flight status information includes the coordinates of the flight start point, the coordinates of the flight end point, and the aircraft parameter information; Based on the flight information, the flight area information corresponding to the flight logistics equipment is generated; Based on the flight area information, the no-fly information that meets the preset conditions is determined as the no-fly area information, thus obtaining the no-fly area information set.

3. The method according to claim 1, wherein, The determination of predicted flight state features corresponding to the flight state vector sequence based on the trajectory optimization model and the no-fly zone information set includes: Based on the Long Short-Term Memory Network and the No-Fly Zone Information Set, temporal features are extracted from the flight state vector sequence to obtain a temporal feature vector set. The last time series feature vector in the set of time series feature vectors is determined as the hidden feature vector; Based on the fully connected layer, feature mapping is performed on the hidden feature vector to obtain the predicted flight state features.

4. A logistics trajectory optimization device for the industrial chain, comprising: The no-fly zone detection unit is configured to detect no-fly zones in the flight area of ​​flying logistics equipment and obtain a set of no-fly zone information. The acquisition unit is configured to acquire a sequence of flight state vectors of the flying logistics equipment. The model training unit is configured to generate a sample training dataset based on numerical methods; Each sample training data in the sample training dataset is preprocessed to generate preprocessed sample training data, resulting in a preprocessed sample training dataset. Each preprocessed sample training data in the preprocessed sample training dataset includes a preprocessed sample state sequence and a preprocessed control quantity true value. The preprocessed sample state sequence included in each preprocessed sample training data in the preprocessed sample training dataset is input into a preset initial trajectory optimization model to obtain a set of predictive control variables. The initial trajectory optimization model includes an initial long short-term memory network, an initial fully connected layer, and an initial neural network. The total loss value is generated based on the loss value between each predicted control variable in the predicted control variable set and the true value of the preprocessed control variable included in the corresponding preprocessed sample training data in the preprocessed sample training dataset. The initial trajectory optimization model is backpropagated and updated based on the total loss value to obtain a pre-trained trajectory optimization model. The pre-trained trajectory optimization model is trained using reinforcement learning to obtain the trajectory optimization model; The execution unit is configured to perform the following trajectory optimization steps in response to the remaining flight range corresponding to the flying logistics equipment being greater than a preset distance value: Based on the trajectory optimization model and the no-fly zone information set, the predicted flight state features corresponding to the flight state vector sequence are determined. The trajectory optimization model includes a long short-term memory network, a fully connected layer, and a neural network of ordinary differential equations. The predicted flight state features are input into the neural network of constant differential equations to obtain optimized control parameters. The flight logistics equipment is controlled according to the optimized control parameters, and the predicted flight state vector corresponding to the optimized control parameters is determined based on the preset flight dynamics equations. Based on the predicted flight state vector and the flight state vector sequence, an updated flight state vector sequence is generated; The second execution unit is configured to execute the trajectory optimization step again using the updated flight state vector sequence as the flight state vector sequence.

5. An electronic device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 3.

6. A computer-readable medium having a computer program stored thereon, wherein, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 3.

Images