CN-122017385-A - Unmanned aerial vehicle electrical appliance state evaluation and fault prediction method, equipment and medium based on ensemble learning

CN122017385ACN 122017385 ACN122017385 ACN 122017385ACN-122017385-A

Abstract

The invention discloses an unmanned aerial vehicle electrical appliance state evaluation and fault prediction method, equipment and medium based on ensemble learning. The method comprises the steps of collecting operation data, denoising by using Daubechies wavelets, retaining transient fault characteristics, searching an optimal LightGBM super-parameter in a super-parameter space by using Bayesian optimization, training a model, outputting a predicted value of the operation parameter at the next moment, calculating a residual error of the predicted value and a true value, and judging abnormality and positioning a fault subsystem when the residual error exceeds the limit. The denoising link is self-adaptive to threshold shrinkage based on the detail coefficient of the finest layer, and the Bayesian link adopts a Gaussian process-TPE acquisition function. The invention can effectively improve the fault positioning accuracy. The full link does not need manual features, supports online dispatching to participate in embedded deployment, and obviously reduces false alarm and operation and maintenance cost.

Inventors

HE CHEN
WANG ZHE
CHEN YIYANG
PU JIAXIN
HUANG JIAN

Assignees

四川腾盾科技有限公司
四川腾盾良远智能科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251223

Claims (10)

1. The unmanned aerial vehicle electrical appliance state evaluation and fault prediction method based on ensemble learning is characterized by comprising the following steps of: acquiring operation data of an unmanned aerial vehicle electrical system, wherein the operation data of the unmanned aerial vehicle electrical system comprises operation parameter data and operation state data; Performing wavelet denoising processing on the operation data of the unmanned aerial vehicle electrical appliance system to obtain denoising data; Inputting the denoising data into a pre-constructed LGBM regression model, and outputting predicted values of all continuous parameters of an unmanned aerial vehicle electrical system, wherein the LGBM regression model is obtained by retraining after searching an optimal super-parameter combination in a super-parameter space through a Bayesian optimization algorithm; and calculating residual errors of the predicted value and the true value at the same moment, and judging that the electrical appliance system is abnormal at the moment if the residual errors exceed a preset threshold value.
2. The method for evaluating and predicting the state of an electrical appliance of an unmanned aerial vehicle based on ensemble learning according to claim 1, wherein the operation parameter data includes one or more of battery voltage, battery current, motor three-phase current, motor rotation speed, motor temperature, controller temperature, battery SOC and battery SOH; the operational status data includes one or more of attitude angle, altitude, GPS signal strength, payload weight, fault codeword, and system self-test flag bit.
3. The method for evaluating the state of an unmanned aerial vehicle electrical appliance and predicting faults based on ensemble learning as claimed in claim 1, wherein the wavelet denoising processing is carried out on the operation data of the unmanned aerial vehicle electrical appliance system, and the method comprises the following steps: selecting Daubechies wavelet basis function to carry out J-layer wavelet decomposition on an original signal, namely the operation data, so as to obtain the approximate coefficient of each layer and the detail coefficient of each layer; calculating a noise standard deviation based on the detail coefficient of the J layer, and determining a noise setting threshold according to the noise standard deviation; Performing soft threshold processing, and performing shrinkage adjustment on the signal coefficient according to the noise set threshold; And carrying out signal reconstruction based on the processed detail coefficient and the approximation coefficient to obtain denoising data of the original signal.
4. The unmanned aerial vehicle electrical appliance state evaluation and fault prediction method based on integrated learning as claimed in claim 1, wherein the LGBM regression model is constructed by a LightGBM framework, an objective function is a mean square error, a basic learner is a gradient lifting tree, model output is an estimated value of continuous parameters of an unmanned aerial vehicle electrical appliance system at the next moment, and an expression of the LGBM regression model is as follows: Wherein, the Representing LGBM a regression model, i.e., a strong learner, x represents the input features; representing an initial model, wherein the regression task is the mean value of training data; Representing an mth weak decision tree; M represents the number of weak decision trees, namely M rounds of iteration are carried out on LGBM regression models; the objective function of the LGBM regression model is expressed as: Wherein, the Y represents a target value; Representing the predicted value.
5. The method for evaluating the state of an unmanned aerial vehicle and predicting faults based on ensemble learning as claimed in claim 4, wherein the LGBM regression model is trained by the following iterative method: in the mth round of iteration, firstly, calculating the negative gradient of the current model to each sample as a training target of a new tree, namely, a pseudo residual error, and comprising: Wherein, the Representing the negative gradient of the ith sample of the mth round, i.e. the pseudo residual; representing a target value corresponding to the ith sample; representing the i-th input feature, i.e., the i-th input sample; representing the strong learner after the m-1 th round of iteration; Solving new tree with minimum tree fitting error as target Is a function of the objective function of: Wherein n represents the total number of training samples; superimposing the predicted result of the new tree into the current model, including: Wherein, the Representing the learning rate.
6. The method for evaluating the state of an unmanned aerial vehicle and predicting faults based on ensemble learning as claimed in claim 1, wherein the searching for the optimal hyper-parameter combination in the hyper-parameter space by a bayesian optimization algorithm comprises: defining a super-parameter vector to be optimized; taking the target function of the Bayesian optimization algorithm as a black box function, establishing a proxy model of the target function of the Bayesian optimization algorithm by using a Gaussian process, and acquiring the prediction distribution of the proxy model; Iteratively updating the proxy model through the acquisition function until the algorithm converges, and outputting an optimal super-parameter combination; And carrying the optimal super parameters into LGBM regression model training flow.
7. The method for evaluating and predicting a state of an unmanned aerial vehicle electrical appliance based on ensemble learning according to claim 6, wherein said creating a proxy model of a bayesian optimization algorithm objective function using a gaussian process and obtaining a proxy model prediction distribution comprises: assuming that the objective function of the bayesian optimization algorithm obeys a gaussian process prior, it is expressed as: Wherein, the Representing the objective function of Bayesian optimization algorithm, m # ) Is a mean function; is a kernel function for measuring two parameter points And Similarity of (2); for unobserved parameters Based on observed data D, non-observed parameters are predicted using a Gaussian process A posterior distribution of (2), comprising: Wherein, the Representing unobserved parameters Is a target function of (2); D represents observed data; Representation pair Is an optimal estimate of (1); representing the uncertainty of the prediction.
8. The method for evaluating and predicting the state of an unmanned aerial vehicle electrical appliance based on ensemble learning of claim 7, wherein said iteratively updating said proxy model by an acquisition function comprises: Selecting super parameters of the next iteration by adopting a TPE acquisition function according to posterior distribution results of unobserved parameters; The TPE acquisition function is expressed as: Wherein, the Representing a TPE acquisition function; Representing probability density of good samples, i.e. the objective function value is greater than the current objective function optimum Is a probability density of a sample of (a); Representing the probability density of bad samples, i.e. the objective function value is not greater than the current objective function optimum Is a probability density of a sample of (a); Selection of The maximum point is used as the super parameter for the next iteration.
9. A computer device comprising a processor and a memory, the memory having stored therein a computer program which when loaded and executed by the processor implements the ensemble learning based unmanned aerial vehicle electrical status assessment and fault prediction method of any one of claims 1 to 8.
10. A computer-readable storage medium, in which a program is stored that, when loaded by a processor, implements the ensemble learning-based unmanned aerial vehicle electrical state evaluation and failure prediction method according to any one of claims 1 to 8.

Description

Unmanned aerial vehicle electrical appliance state evaluation and fault prediction method, equipment and medium based on ensemble learning Technical Field The invention relates to the technical field of unmanned aerial vehicle health monitoring, in particular to an unmanned aerial vehicle electrical appliance state evaluation and fault prediction method, equipment and medium based on ensemble learning. Background Along with the wide application of unmanned aerial vehicles in the fields of electric power inspection, logistics transportation, emergency rescue and the like, the reliability of an electric system of the unmanned aerial vehicle has become a core element for guaranteeing flight safety. The components such as the battery, the motor, the electric control and the like run for a long time under high-altitude strong electromagnetic, severe maneuvering and wide-temperature environments, and early faults such as insulation aging, abnormal commutation, over-temperature and the like are extremely easy to occur. However, signals such as voltage, current, temperature and the like collected by the unmanned aerial vehicle-mounted sensor are often accompanied by high-frequency pulses and random drift, and the traditional low-pass filtering or moving average method can weaken transient fault pulses while suppressing noise, so that early fault characteristics are submerged, and follow-up diagnosis models are missed. In terms of state evaluation and fault prediction, shallow models such as threshold overrun alarm or support vector machines, random forests and the like are commonly adopted in engineering. The method has poor adaptability to high-dimensional, sparse and nonlinear unmanned aerial vehicle electrical appliance data, relies on artificial characteristic engineering, and the model precision is rapidly reduced along with the change of working conditions. In recent years, the gradient lifting tree (GBDT) and the variation LightGBM thereof show the advantages of high precision and high speed in a high-dimensional regression task, but the hyper-parameter selection still depends on grid search or empirical debugging, is time-consuming in calculation and is easy to sink into local optimum, and is difficult to meet the aging requirement of unmanned aerial vehicle online maintenance. Bayesian optimization, which guides the super-parameter search through a gaussian process proxy model, can significantly reduce the number of evaluation times, has been used for XGBoost and neural network tuning. However, bayesian optimization and LightGBM are combined, systematic optimization is carried out on small-sample and high-noise unmanned aerial vehicle electrical appliance data, the prior literature is not reported, meanwhile, the correlation between an electrical appliance signal frequency band and a fault mechanism is not considered in the prior research, weak fault components are easily lost in the denoising process, and the subsequent model recall rate is insufficient. Therefore, a systematic scheme of integrating 'feature preserving denoising-super parameter optimization-regression prediction' is needed to improve the accuracy of unmanned aerial vehicle electrical system state evaluation and the timeliness of fault early warning. Disclosure of Invention The present invention aims to solve at least one of the above technical problems in the prior art. Therefore, the first aspect of the invention provides an unmanned aerial vehicle electrical appliance state evaluation and fault prediction method based on ensemble learning. A second aspect of the invention provides a computer device. A third aspect of the present invention provides a computer-readable storage medium. The invention provides an unmanned aerial vehicle electrical appliance state evaluation and fault prediction method based on ensemble learning, which comprises the following steps: acquiring operation data of an unmanned aerial vehicle electrical system, wherein the operation data of the unmanned aerial vehicle electrical system comprises operation parameter data and operation state data; Performing wavelet denoising processing on the operation data of the unmanned aerial vehicle electrical appliance system to obtain denoising data; Inputting the denoising data into a pre-constructed LGBM regression model, and outputting predicted values of all continuous parameters of an unmanned aerial vehicle electrical system, wherein the LGBM regression model is obtained by retraining after searching an optimal super-parameter combination in a super-parameter space through a Bayesian optimization algorithm; and calculating residual errors of the predicted value and the true value at the same moment, and judging that the electrical appliance system is abnormal at the moment if the residual errors exceed a preset threshold value. According to the technical scheme, the unmanned aerial vehicle electrical appliance state evaluation and fault prediction method based on the in