CN113342020A - Method for predicting propulsion power of electric propulsion unmanned aerial vehicle - Google Patents

Method for predicting propulsion power of electric propulsion unmanned aerial vehicle Download PDF

Info

Publication number
CN113342020A
CN113342020A CN202110657753.XA CN202110657753A CN113342020A CN 113342020 A CN113342020 A CN 113342020A CN 202110657753 A CN202110657753 A CN 202110657753A CN 113342020 A CN113342020 A CN 113342020A
Authority
CN
China
Prior art keywords
unmanned aerial
aerial vehicle
neural network
electric propulsion
deep neural
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110657753.XA
Other languages
Chinese (zh)
Inventor
雷涛
闵志豪
张星雨
王彦博
张晓斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northwestern Polytechnical University
Original Assignee
Northwestern Polytechnical University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Northwestern Polytechnical University filed Critical Northwestern Polytechnical University
Priority to CN202110657753.XA priority Critical patent/CN113342020A/en
Publication of CN113342020A publication Critical patent/CN113342020A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course or altitude of land, water, air, or space vehicles, e.g. automatic pilot
    • G05D1/08—Control of attitude, i.e. control of roll, pitch, or yaw
    • G05D1/0808—Control of attitude, i.e. control of roll, pitch, or yaw specially adapted for aircraft
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course or altitude of land, water, air, or space vehicles, e.g. automatic pilot
    • G05D1/10—Simultaneous control of position or course in three dimensions
    • G05D1/101—Simultaneous control of position or course in three dimensions specially adapted for aircraft

Abstract

The invention discloses a prediction method of propulsion power of an electric propulsion unmanned aerial vehicle, which comprises the steps of firstly designing an airborne data acquisition system of a distributed electric propulsion unmanned aerial vehicle, carrying out data acquisition on a plurality of flight tests of the distributed full electric propulsion unmanned aerial vehicle taking a lithium battery as a power source under different working conditions and different environments, and preprocessing the acquired data to obtain a data set required by model training and prediction; secondly, establishing a deep BP neural network, and respectively training and testing the model by using the obtained data set; and finally, analyzing the prediction effect of the model. The invention trains by taking a plurality of groups of real data for flight tests under different flight conditions and flight environments as a training set, can predict the propulsion power of the unmanned aerial vehicle under different conditions and different flight environments, and has great engineering and theoretical significance for precise energy management and load side management.

Description

Method for predicting propulsion power of electric propulsion unmanned aerial vehicle
Technical Field
The invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to a power prediction method for an unmanned aerial vehicle.
Background
Most of the existing energy management strategies for electric propulsion power systems are researched under a certain pre-known fixed working condition, so that although a certain index of the power system under the working condition can be improved after energy management, the strategy adaptability and robustness under the conditions of variable working conditions and uncertain operating environments are still insufficient. The performance index of the energy management strategy in this aspect can be improved by combining with real-time load prediction. Currently, a great deal of research has been conducted on load prediction for ground grid systems. The british national grid has conducted research on weather-related variables to predict the peak value of british power supply and demand, which is expected to reduce the operating cost of the grid through accurate load prediction.
The methods for load prediction can be classified according to conventional methods as well as intelligent methods. The conventional prediction methods include a multiple linear regression method, an exponential smoothing method, and a differential autoregressive moving average method. However, due to the non-linear characteristic of the load, the traditional methods are difficult to achieve the ideal prediction effect. The intelligent prediction method has great advantages for the problems of nonlinear learning and modeling. The intelligent prediction method comprises a clustering method, a fuzzy logic system, a Support Vector Machine (SVM) and an artificial neural network. Deep learning possesses multiple hidden layers compared to shallow learning, which enables neural networks to learn more complex nonlinear models. Recurrent neural networks are able to capture non-stationary and long-term correlated prediction ranges, which are typically used to make grid load predictions. In the prior art, a novel pool-based deep RNN is applied to load prediction of families and has been primarily successful; a new load prediction model introduces further look-ahead concepts into the RNN model. However, gradient explosion is a problem that restricts RNN performance, and compared with a ground power grid, flight thrust load and task load of a distributed electric propulsion aircraft generate uncertain changes in a flight envelope due to environmental influences, so that training and learning of test flight data by using a deep neural network can effectively solve the problem.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a prediction method of the propulsion power of an electric propulsion unmanned aerial vehicle, which comprises the steps of firstly designing an airborne data acquisition system of the distributed electric propulsion unmanned aerial vehicle, acquiring data of a plurality of flight tests of the distributed full electric propulsion unmanned aerial vehicle taking a lithium battery as a power source under different working conditions and different environments, and preprocessing the acquired data to obtain a data set required by model training and prediction; secondly, establishing a deep BP neural network, and respectively training and testing the model by using the obtained data set; and finally, analyzing the prediction effect of the model. The invention trains by taking a plurality of groups of real data for flight tests under different flight conditions and flight environments as a training set, can predict the propulsion power of the unmanned aerial vehicle under different conditions and different flight environments, and has great engineering and theoretical significance for precise energy management and load side management.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
step 1: collecting training data;
installing an airborne sensor on the distributed electric propulsion unmanned aerial vehicle, and acquiring data required by training through flight tests of the distributed electric propulsion unmanned aerial vehicle under different flight working conditions and different flight activities, wherein the data comprises the height, the track angle, the attack angle, the airspeed, the voltage and the current of the distributed electric propulsion unmanned aerial vehicle;
carrying out moving average filtering on the acquired voltage and current values to filter out singular points; calculating to obtain the electric power demand of the electric propulsion unmanned aerial vehicle at the corresponding height, track angle, attack angle and airspeed of the voltage and the current;
step 2: predicting the electric power of the electric propulsion unmanned aerial vehicle by adopting a BP deep neural network;
the input of the BP deep neural network is the height, the track angle, the attack angle and the airspeed of the distributed electric propulsion unmanned aerial vehicle, and is expressed as an input vector X(i)＝[h(i),γ(i),α(i),v(i)]TWherein h, gamma, alpha and v are respectively the height, track angle, attack angle and airspeed of the distributed electric propulsion unmanned aerial vehicle; the output of the BP deep neural network is electric power of the electric propulsion unmanned aerial vehicle, and is expressed as
Figure BDA0003113991080000021
The root mean square error RMSE and the absolute value percentage error MAPE are selected as loss functions, the formula for calculating RMSE is shown as a formula (3), and the formula for calculating MAPE is shown as a formula (4).
Figure BDA0003113991080000022
Figure BDA0003113991080000023
In the formula, ytRepresenting the output value of the neural network model at time t, ftAt the time T, the observed value of the neural network model, wherein T is the number of data points;
training the BP deep neural network by adopting the training data obtained in the step 1, and updating the weight of the BP deep neural network to obtain a final BP deep neural network;
and step 3: and acquiring the height, track angle, attack angle and airspeed of the distributed electric propulsion unmanned aerial vehicle in real time, inputting the final BP deep neural network, and acquiring the predicted value of the electric power of the electric propulsion unmanned aerial vehicle.
Preferably, the on-board sensors include pitot tube, dynamic pressure sensors, GPS sensors, voltage and current sensors.
Preferably, when the BP deep neural network is trained, the learning rate is set to 0.005;
preferably, each hidden layer of the BP deep neural network is designed to be 10 neurons, and the number of the hidden layer layers is 5.
The invention has the following beneficial effects:
the invention trains by taking a plurality of groups of real data for flight tests under different flight conditions and flight environments as a training set, can predict the propulsion power of the unmanned aerial vehicle under different conditions and different flight environments, and has great engineering and theoretical significance for precise energy management and load side management. Compared with the traditional energy optimization management method designed under a certain specific flight environment and flight condition determined task profile, the method for predicting the propulsion power based on the neural network can predict the propulsion power demand in real time through a plurality of groups of prediction models trained by flight experimental data under different conditions and flight environments, so that the energy optimization management strategy has better robustness and adaptivity.
Drawings
FIG. 1 is a study roadmap of an electric propulsion power prediction model according to the present invention.
Fig. 2 is a flight deck propulsion power curve according to the present invention.
FIG. 3 is a propulsion power prediction model structure based on a deep neural network.
FIG. 4 is the deep neural network training set input data of the present invention.
FIG. 5 is the expected output data of the deep neural network training set of the present invention.
FIG. 6 is the deep neural network training results of the present invention.
Fig. 7 shows the prediction results of different iteration steps of the present invention, where (a) is the result of 50-step iteration prediction of the deep BP network, (b) is the result of 100-step iteration prediction of the deep BP network, (c) is the result of 200-step iteration prediction of the deep BP network, (d) is the result of 500-step iteration prediction of the deep BP network, and (e) is the result of 1000-step iteration prediction of the deep BP network.
FIG. 8 is a predicted result of a flight number propulsion power demand according to an embodiment of the present invention.
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
For the power system of the distributed electric propulsion unmanned aerial vehicle, there are many factors that affect the flight speed of the unmanned aerial vehicle, such as track, flight attitude, airspeed, wind speed, and the like. The distributed model of the electric propulsion unmanned aerial vehicle is a strong nonlinear model, so that a deep neural network is selected to model a propulsion power demand model of the electric propulsion unmanned aerial vehicle, and the model is tested.
As shown in fig. 1, a method for predicting the propulsion power of an electric propulsion unmanned aerial vehicle includes the following steps:
step 1: collecting training data;
installing an airborne sensor on the distributed electric propulsion unmanned aerial vehicle, wherein the airborne sensor comprises an airspeed head, a dynamic pressure sensor, a GPS sensor and a voltage and current sensor; acquiring data required by training through flight tests of a plurality of unmanned aerial vehicles under different flight working conditions and different flight activities, wherein the data comprises the height, track angle, attack angle, airspeed, voltage and current of the distributed electric propulsion unmanned aerial vehicle;
carrying out moving average filtering on the acquired voltage and current values to filter out singular points; calculating to obtain the electric power demand of the electric propulsion unmanned aerial vehicle at the corresponding height, track angle, attack angle and airspeed of the voltage and the current;
step 2: predicting the electric power of the electric propulsion unmanned aerial vehicle by adopting a BP deep neural network;
the input of the BP deep neural network is the height, the track angle, the attack angle and the airspeed of the distributed electric propulsion unmanned aerial vehicle, and is expressed as an input vector X(i)＝[h(i),γ(i),α(i),v(i)]TWherein h, gamma, alpha and v are respectively the height, track angle, attack angle and airspeed of the distributed electric propulsion unmanned aerial vehicle; the output of the BP deep neural network is electric power of the electric propulsion unmanned aerial vehicle, and is expressed as
Figure BDA0003113991080000041
The root mean square error RMSE and the absolute value percentage error MAPE are selected as loss functions, the formula for calculating RMSE is shown as a formula (3), and the formula for calculating MAPE is shown as a formula (4).
Figure BDA0003113991080000042
Figure BDA0003113991080000043
In the formula, ytRepresenting the output value of the neural network model at time t, ftAt the time T, the observed value of the neural network model, wherein T is the number of data points;
training the BP deep neural network by adopting the training data obtained in the step 1, and updating the weight of the BP deep neural network to obtain a final BP deep neural network;
and step 3: and acquiring the height, track angle, attack angle and airspeed of the distributed electric propulsion unmanned aerial vehicle in real time, inputting the final BP deep neural network, and acquiring the predicted value of the electric power of the electric propulsion unmanned aerial vehicle.
The specific embodiment is as follows:
1. acquiring a training data set;
through flight test under the different flight tracks of many times, different environment, the design installation test sensor is installed on distributed electric propulsion unmanned aerial vehicle, gathers the required parameter data of training when flight test. The sensors mainly comprise an airspeed head, a dynamic pressure sensor, a GPS sensor and a voltage and current sensor.
When the distributed electric propulsion unmanned aerial vehicle carries out flight tests, flight frames under different flight working conditions and different flight activities are subjected to data acquisition through an airborne sensor and an acquisition system, training data sets are accumulated, and the data sets used for training can be expanded by increasing the acquired data of different frames; the flight conditions and environmental conditions for the partial rack are shown in table 1.
TABLE 1 flight test conditions and Environment
Figure BDA0003113991080000051
After the acquired voltage and current signals are obtained, the electric power requirement of the electric propulsion unmanned aerial vehicle can be calculated. Because the sampling of the test system has errors, the obtained current and voltage data are preprocessed by adopting the moving average filtering, and the singular points of the sampling are filtered. The resulting electrical propulsion power requirements for a flight deck after processing are shown in fig. 2.
2. Deep neural network model
Deep Neural Networks (DNNs) are understood to be Neural Networks with many hidden layers. The deep neural network has a main structure as shown in fig. 3, and includes an input layer, an output layer, and a plurality of hidden layers.
For a single artificial neuron at each layer, the model has three elements: the device comprises a connection weight, a summation unit of input information after the weight is acted, and an activation function.
The connection weight can reflect the relationship between connected neurons, and if the relationship is positive, activation is indicated, and otherwise, inhibition is indicated. The role of the activation function in the neural network is to perform weighted sum of input vectors and perform nonlinear mapping, and to limit the output amplitude of the artificial neuron within a certain range, and the types of the activation function are many, such as sigmoid, tanh, and the like. The mathematical relationship of a single neuron is shown in formulas (3) and (4).
Figure BDA0003113991080000052
yk＝φ(xk+θk) (4)
The BP algorithm is a method based on the fastest gradient descent, and the mean square error of the output predicted value and the expected value of the neural network is minimized by adjusting the weight by using a mathematical method of gradient search.
When the deep neuron network learns by applying a BP algorithm, input information is processed layer by layer from an input layer through a hidden layer and is transmitted to an output layer. The condition of entering the backward propagation is that the expected output can not be obtained, the error signal is transmitted back according to the original propagation path, and then the weight of each layer of neuron is modified, thereby achieving the effect of minimizing the error signal. The calculation steps of the BP algorithm are as follows:
TABLE 2 deep neural network BP algorithm procedure
Figure BDA0003113991080000061
For a deep neural network model for prediction of the propulsion power demand of an electrically propelled aircraft, altitude, track angle, angle of attack, airspeed may be selected as input vector X(i)＝[h(i),γ(i),α(i),v(i)]ΤThe desired output vector is selected as the propulsion power demand, i.e.
Figure BDA0003113991080000062
Through data training, the weight of the neural network is updated, and therefore the effect of predicting the propulsion power demand is achieved.
3. Predictive model training
A plurality of groups of data obtained by a plurality of test flights are respectively used as data sets of the deep neural network for training, and the data sets for training and testing are shown in fig. 4 and 5. The first 86% of the data set sequence is set as the training set and the last 14% is set as the test set. After debugging, the learning rates of the two models are set to be 0.005, each hidden layer of the deep neural network is designed to be 10 neurons, and the number of the hidden layers is 5. After parameter design is completed, training the network by using the training set, dividing the test set from the sample according to a certain proportion, and verifying the prediction effect by using the test set, wherein the training effect is shown in fig. 6.
4. Predictive model testing
When a neural network algorithm is used for load or load power demand prediction, a loss function is usually used to express the accuracy of a prediction model so as to evaluate the prediction effect. Selecting Root Mean Square Error (RMSE) and Absolute value Percentage Error (MAPE) as loss functions, and respectively calculating the values of RMSE and MAPE for the training set and the test set so as to realize the evaluation of the accuracy of the prediction model.
Setting two prediction models to train with the same number of neurons, the same learning rate, the same data set and the same iteration times; the training effect is compared by changing the iteration number. Two loss functions of the method at different iteration times are shown in the table 3, and a predicted propulsion power demand curve and a real propulsion power curve are shown in the fig. 7.
TABLE 3 two loss functions for different iteration steps
Figure BDA0003113991080000071
Through comparative analysis, the parameters of the established prediction model are determined as shown in table 4.
TABLE 4 selected prediction model parameter settings
Figure BDA0003113991080000072
The method is realized by programming in Matlab, the propulsion power is predicted by adopting flight data acquired by an electric propulsion unmanned aerial vehicle for one flight frame, the prediction result is shown in figure 8, the mean square error value is 0.0205, the propulsion power prediction method based on the deep neural network has high prediction precision, and the propulsion power requirements of the unmanned aerial vehicle under different flight conditions can be predicted.

Claims (4)

1. The method for predicting the propulsion power of the electric propulsion unmanned aerial vehicle is characterized by comprising the following steps of:
step 1: collecting training data;
installing an airborne sensor on the distributed electric propulsion unmanned aerial vehicle, and acquiring data required by training through flight tests of the distributed electric propulsion unmanned aerial vehicle under different flight working conditions and different flight activities, wherein the data comprises the height, the track angle, the attack angle, the airspeed, the voltage and the current of the distributed electric propulsion unmanned aerial vehicle;
carrying out moving average filtering on the acquired voltage and current values to filter out singular points; calculating to obtain the electric power demand of the electric propulsion unmanned aerial vehicle at the corresponding height, track angle, attack angle and airspeed of the voltage and the current;
step 2: predicting the electric power of the electric propulsion unmanned aerial vehicle by adopting a BP deep neural network;
the input of the BP deep neural network is a distributed electric propulsion unmanned aerial vehicleAltitude, track angle, angle of attack, airspeed, expressed as an input vector X(i)＝[h(i),γ(i),α(i),v(i)]TWherein h, gamma, alpha and v are respectively the height, track angle, attack angle and airspeed of the distributed electric propulsion unmanned aerial vehicle; the output of the BP deep neural network is electric power of the electric propulsion unmanned aerial vehicle, and is expressed as
Figure FDA0003113991070000011
The root mean square error RMSE and the absolute value percentage error MAPE are selected as loss functions, the formula for calculating RMSE is shown as a formula (3), and the formula for calculating MAPE is shown as a formula (4).
Figure FDA0003113991070000012
Figure FDA0003113991070000013
In the formula, ytRepresenting the output value of the neural network model at time t, ftAt the time T, the observed value of the neural network model, wherein T is the number of data points;
training the BP deep neural network by adopting the training data obtained in the step 1, and updating the weight of the BP deep neural network to obtain a final BP deep neural network;
and step 3: and acquiring the height, track angle, attack angle and airspeed of the distributed electric propulsion unmanned aerial vehicle in real time, inputting the final BP deep neural network, and acquiring the predicted value of the electric power of the electric propulsion unmanned aerial vehicle.
2. The method of claim 1, wherein the onboard sensors comprise airspeed head, dynamic pressure sensor, GPS sensor, voltage and current sensor.
3. The method of claim 1, wherein the learning rate is set to 0.005 when the BP deep neural network is trained.
4. The method of claim 1, wherein each hidden layer of the BP deep neural network is designed to be 10 neurons, and the number of hidden layer layers is 5.
CN202110657753.XA 2021-06-14 2021-06-14 Method for predicting propulsion power of electric propulsion unmanned aerial vehicle Pending CN113342020A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110657753.XA CN113342020A (en) 2021-06-14 2021-06-14 Method for predicting propulsion power of electric propulsion unmanned aerial vehicle

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110657753.XA CN113342020A (en) 2021-06-14 2021-06-14 Method for predicting propulsion power of electric propulsion unmanned aerial vehicle

Publications (1)

Publication Number Publication Date
CN113342020A true CN113342020A (en) 2021-09-03

Family

ID=77476883

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110657753.XA Pending CN113342020A (en) 2021-06-14 2021-06-14 Method for predicting propulsion power of electric propulsion unmanned aerial vehicle

Country Status (1)

Country Link
CN (1) CN113342020A (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110348595A (en) * 2019-05-31 2019-10-18 南京航空航天大学 A kind of unmanned plane mixed propulsion system energy management-control method based on flying quality
CN111703580A (en) * 2020-05-28 2020-09-25 上海交通大学 Electric propulsion rotor craft power system and control method thereof
CN112925344A (en) * 2021-01-25 2021-06-08 南京航空航天大学 Unmanned aerial vehicle flight condition prediction method based on data driving and machine learning

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110348595A (en) * 2019-05-31 2019-10-18 南京航空航天大学 A kind of unmanned plane mixed propulsion system energy management-control method based on flying quality
CN111703580A (en) * 2020-05-28 2020-09-25 上海交通大学 Electric propulsion rotor craft power system and control method thereof
CN112925344A (en) * 2021-01-25 2021-06-08 南京航空航天大学 Unmanned aerial vehicle flight condition prediction method based on data driving and machine learning

Similar Documents

Publication Publication Date Title
CN111830408B (en) Motor fault diagnosis system and method based on edge calculation and deep learning
CN108594788B (en) Airplane actuator fault detection and diagnosis method based on deep random forest algorithm
CN112131760B (en) CBAM model-based prediction method for residual life of aircraft engine
CN106779151B (en) A kind of line of high-speed railway wind speed multi-point multi-layer coupling prediction method
CN109308522B (en) GIS fault prediction method based on recurrent neural network
CN109141847B (en) Aircraft system fault diagnosis method based on MSCNN deep learning
CN109766583A (en) Based on no label, unbalanced, initial value uncertain data aero-engine service life prediction technique
WO2015158198A1 (en) Fault recognition method and system based on neural network self-learning
CN107817404B (en) Portable metering automation terminal fault diagnosis device and diagnosis method thereof
CN111428942B (en) Line icing thickness prediction method for extracting micro-terrain factors based on variable grid technology
CN111680875B (en) Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model
CN111461413B (en) Highway road surface performance detecting system
CN110555551A (en) air quality big data management method and system for smart city
CN115220133B (en) Rainfall prediction method, device and equipment for multiple meteorological elements and storage medium
CN115577860A (en) Intelligent maintenance method and system for rail transit based on adaptive control
CN116796168A (en) CNN-BiLSTM high-altitude multi-factor power transmission line audible noise prediction method based on multi-head attention mechanism
CN113076686A (en) Aircraft trajectory prediction method based on social long-term and short-term memory network
CN113342020A (en) Method for predicting propulsion power of electric propulsion unmanned aerial vehicle
JPH0949884A (en) Predicting method for local weather using neural network
CN113108949B (en) Model fusion-based sonde temperature sensor error prediction method
CN115017984A (en) Early warning method and system for failure risk of aircraft engine
CN114372640A (en) Wind power prediction method based on fluctuation sequence classification correction
Kou Wireless communication system and its application in big data remote monitoring and decision-making
CN115618988A (en) Three-dimensional space-time field joint prediction method for seawater temperature, salinity and flow velocity based on deep neural network
CN112560252A (en) Prediction method for residual life of aircraft engine

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20210903