CN118070204A - Unmanned aerial vehicle power data anomaly identification method and device based on neural network - Google Patents
Unmanned aerial vehicle power data anomaly identification method and device based on neural network Download PDFInfo
- Publication number
- CN118070204A CN118070204A CN202410487195.0A CN202410487195A CN118070204A CN 118070204 A CN118070204 A CN 118070204A CN 202410487195 A CN202410487195 A CN 202410487195A CN 118070204 A CN118070204 A CN 118070204A
- Authority
- CN
- China
- Prior art keywords
- data
- unmanned aerial
- aerial vehicle
- neural network
- quantum
- 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.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 112
- 238000000034 method Methods 0.000 title claims abstract description 78
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims description 97
- 230000006870 function Effects 0.000 claims description 61
- 230000002159 abnormal effect Effects 0.000 claims description 47
- 238000004422 calculation algorithm Methods 0.000 claims description 34
- 239000013598 vector Substances 0.000 claims description 34
- 239000011159 matrix material Substances 0.000 claims description 24
- 230000004913 activation Effects 0.000 claims description 23
- 230000008859 change Effects 0.000 claims description 21
- 238000012544 monitoring process Methods 0.000 claims description 20
- 238000000605 extraction Methods 0.000 claims description 18
- 238000005457 optimization Methods 0.000 claims description 18
- 210000002569 neuron Anatomy 0.000 claims description 11
- 238000005070 sampling Methods 0.000 claims description 11
- 238000005516 engineering process Methods 0.000 claims description 9
- 230000005428 wave function Effects 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 230000003044 adaptive effect Effects 0.000 claims description 6
- 230000015572 biosynthetic process Effects 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 6
- 238000003786 synthesis reaction Methods 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 5
- 230000002776 aggregation Effects 0.000 claims description 4
- 238000004220 aggregation Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 2
- 230000002194 synthesizing effect Effects 0.000 claims 2
- 230000005856 abnormality Effects 0.000 abstract description 15
- 230000000875 corresponding effect Effects 0.000 description 32
- 238000010586 diagram Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 238000000137 annealing Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 238000004590 computer program Methods 0.000 description 4
- 239000000306 component Substances 0.000 description 3
- 239000002131 composite material Substances 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 230000002093 peripheral effect Effects 0.000 description 3
- 239000002096 quantum dot Substances 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 241000282414 Homo sapiens Species 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000003416 augmentation Effects 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 210000002364 input neuron Anatomy 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013021 overheating Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000010363 phase shift Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method and a device for recognizing abnormality of unmanned aerial vehicle power data based on a neural network, which relate to the technical field of data processing. In addition, the invention also converts the recognition result of the model into the confidence score, so that the classification result of the model can be evaluated more accurately, and the recognition result is more reliable.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a neural network-based unmanned aerial vehicle power data anomaly identification method and device.
Background
With the rapid development of unmanned aerial vehicle technology and the wide application of unmanned aerial vehicle technology in the fields of military reconnaissance, geographical exploration, disaster monitoring, logistics distribution and the like, the safe operation of unmanned aerial vehicles becomes an important subject of research and practice. The unmanned aerial vehicle power system is used as one of the core components, and the reliability of the performance of the unmanned aerial vehicle power system is directly related to the safety and efficiency of the unmanned aerial vehicle. However, in a complex flight environment, the power system is susceptible to various internal and external factors, and abnormal conditions may occur, such as overheating of a motor, performance degradation of a battery, thrust loss and the like, and if the abnormal conditions are not found and handled in time, unmanned aerial vehicles may be out of control or even crash, resulting in property loss and safety accidents.
Conventional unmanned aerial vehicle power data anomaly identification techniques typically employ threshold-based, statistical-based, and model-based methods. These methods are better in handling simple abnormal data, but less effective in handling complex abnormal data. With the development of big data and machine learning techniques, researchers have begun to attempt to use these techniques to deal with anomalies in unmanned power data. However, due to the specificity of the unmanned power data, the existing machine learning model is not suitable for the identification of the unmanned power data anomalies.
Disclosure of Invention
In view of the above, the invention aims to provide a neural network-based unmanned aerial vehicle power data abnormality identification method and device, which can better identify unmanned aerial vehicle power data abnormality and improve identification accuracy.
In a first aspect, an embodiment of the present invention provides a method for identifying an abnormality in power data of an unmanned aerial vehicle based on a neural network, where the method includes: acquiring power data of an unmanned aerial vehicle to be identified; extracting features of the unmanned aerial vehicle power data, and determining key features in the unmanned aerial vehicle power data; identifying key features through a pre-constructed data identification model, and determining an identification result corresponding to the unmanned aerial vehicle power data; the training sample set of the training data recognition model comprises a training sample and a sample label, wherein the sample label is used for representing the abnormal condition corresponding to the training sample; the training sample is constructed based on data of a plurality of monitoring points of the unmanned aerial vehicle power system, and data expansion is carried out by adopting quantum coding and quantum gates; converting the identification result into a confidence score; and when the confidence score meets a preset confidence threshold, determining the abnormal condition corresponding to the unmanned aerial vehicle power data based on the identification result.
In a second aspect, an embodiment of the present invention provides an apparatus for identifying abnormality in power data of an unmanned aerial vehicle based on a neural network, where the apparatus includes: the data acquisition module is used for acquiring power data of the unmanned aerial vehicle to be identified; the data processing module is used for extracting the characteristics of the unmanned aerial vehicle power data and determining key characteristics in the unmanned aerial vehicle power data; the execution module is used for identifying key features through a pre-constructed data identification model and determining an identification result corresponding to the unmanned aerial vehicle power data; the training sample set of the training data recognition model comprises a training sample and a sample label, wherein the sample label is used for representing the abnormal condition corresponding to the training sample; the training sample is constructed based on data of a plurality of monitoring points of the unmanned aerial vehicle power system, and data expansion is carried out by adopting quantum coding and quantum gates; the data conversion module converts the identification result into a confidence score; and the output module is used for determining abnormal conditions corresponding to the unmanned aerial vehicle power data based on the recognition result when the confidence score meets a preset confidence threshold.
The embodiment of the invention has the following beneficial effects: according to the unmanned aerial vehicle power data anomaly identification method and device based on the neural network, after the unmanned aerial vehicle power data are obtained, the data are subjected to feature extraction, and the data are identified by the pre-constructed data identification model, wherein a training sample set for training the model is constructed based on the data of a plurality of monitoring points of an unmanned aerial vehicle power system, and data expansion is performed by adopting quantum coding and quantum gates, so that the training sample set has diversity, the unmanned aerial vehicle power data anomaly can be better identified, and the identification precision is improved. In addition, the embodiment of the invention also converts the recognition result of the model into the confidence score, so that the classification result of the model can be evaluated more accurately, and the recognition result is more reliable.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings. In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of a method for identifying abnormal power data of an unmanned aerial vehicle based on a neural network according to an embodiment of the present invention;
Fig. 2 is a flowchart of another method for identifying abnormal power data of an unmanned aerial vehicle based on a neural network according to an embodiment of the present invention;
Fig. 3 is a flowchart of a third method for identifying abnormal power data of an unmanned aerial vehicle based on a neural network according to an embodiment of the present invention;
Fig. 4 is a flowchart of a fourth method for identifying abnormal power data of an unmanned aerial vehicle based on a neural network according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an abnormal recognition device for power data of an unmanned aerial vehicle based on a neural network according to an embodiment of the present invention;
Fig. 6 is a schematic structural diagram of another abnormal recognition device for power data of an unmanned aerial vehicle based on a neural network according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
For the purposes of clarity, technical solutions, and advantages of the embodiments of the present disclosure, the following description describes embodiments of the present disclosure with specific examples, and other advantages and effects of the present disclosure will be readily apparent to those skilled in the art from the disclosure herein. It will be apparent that the described embodiments are merely some, but not all embodiments of the present disclosure. The disclosure may be embodied or practiced in other different specific embodiments, and details within the subject specification may be modified or changed from various points of view and applications without departing from the spirit of the disclosure. It should be noted that the following embodiments and features in the embodiments may be combined with each other without conflict. All other embodiments, which can be made by one of ordinary skill in the art without inventive effort, based on the embodiments in this disclosure are intended to be within the scope of this disclosure.
The unmanned aerial vehicle power data abnormality identification method and device based on the neural network provided by the embodiment of the invention can better identify the unmanned aerial vehicle power data abnormality and improve the identification precision.
Example 1
In order to facilitate understanding, firstly, a detailed description is given of a neural network-based method for identifying abnormal power data of an unmanned aerial vehicle provided by the embodiment of the present invention, fig. 1 shows a flowchart of the neural network-based method for identifying abnormal power data of an unmanned aerial vehicle provided by the embodiment of the present invention, and as shown in fig. 1, the method includes the following steps:
step S102, acquiring power data of the unmanned aerial vehicle to be identified.
When the method is specifically implemented, data acquisition is firstly carried out on the unmanned aerial vehicle power system, wherein the data of the embodiment of the invention is derived from the unmanned aerial vehicle power system and comprises data for acquiring a plurality of monitoring points of the unmanned aerial vehicle power system. Wherein each monitoring point has data of a plurality of attributes. Each attribute corresponds to a different data characteristic.
Specifically, each piece of data is a tuple, which can be expressed as: d= (t, p 1,p2,…,pm); wherein t is a time stamp representing a time point of data acquisition; p i is the data of monitoring points i, wherein the data p i of each monitoring point comprises data of voltage, current, temperature, humidity, wind speed and the like, and each data is characterized by the numerical value of attribute data thereof; m is the total number of monitoring points.
And step S104, extracting the characteristics of the unmanned aerial vehicle power data, and determining key characteristics in the unmanned aerial vehicle power data.
Further, key features in the unmanned aerial vehicle power data are extracted, and in the specific implementation, the embodiment of the invention performs feature extraction through a feature extraction model so as to determine key information in the data.
And S106, identifying key features through a pre-constructed data identification model, and determining an identification result corresponding to the unmanned aerial vehicle power data.
In specific implementation, the embodiment of the invention utilizes the trained model to perform the task of identifying and classifying the abnormal power data of the unmanned aerial vehicle, and evaluates or verifies the classification result. For inputting one unmanned power data, the model will output the category to which the data belongs. Specifically, let theRepresenting the input samples, the output of the model is/>The course of reasoning can be expressed as:
wherein, Is a trained model. /(I)A probability vector is output for the classification model, each element of which represents the probability of a respective class.
Step S108, converting the identification result into a confidence score.
And step S110, determining abnormal conditions corresponding to the unmanned aerial vehicle power data based on the recognition result when the confidence score meets a preset confidence threshold.
In the embodiment of the invention, the probability vector (i.e. the recognition result) output by the model can be converted into a confidence score for evaluating the classification result. Specifically, let theIndicating that the sample belongs to the/>Probability of individual category, confidence score/>The calculation can be made by the following formula:
wherein, Is the total number of categories.
Further, the embodiment of the invention sets a threshold valueThe classification result can be evaluated more accurately. If the confidence score of a certain category is greater than or equal to the threshold/>The recognition classification result is considered to be reliable.
In specific implementation, a training sample set for training the data identification model of the embodiment of the invention comprises a training sample and a sample label, wherein the sample label is used for representing an abnormal condition corresponding to the training sample; the training sample is constructed based on data of a plurality of monitoring points of the unmanned aerial vehicle power system, and data expansion is carried out by adopting quantum coding and quantum gates. The training sample set has diversity, can better identify the abnormality of the power data of the unmanned aerial vehicle, and improves the identification precision. In addition, the embodiment of the invention also converts the recognition result of the model into the confidence score, so that the classification result of the model can be evaluated more accurately, and the recognition result is more reliable.
Example two
Further, the prior art has the problems of data expansion, feature extraction and classifier training, and based on the problems, the prior art has lower accuracy in the aspect of unmanned aerial vehicle power data anomaly identification, and is insufficient for meeting the requirements in practical application.
Based on the problems, the embodiment of the invention also provides another unmanned aerial vehicle power data anomaly identification method based on the neural network on the basis of the embodiment, and the method mainly describes the construction process of the data identification model. The embodiment of the invention constructs a data identification model through a preset training sample set, and can perform feature extraction on the training sample set through a pre-constructed feature extraction model when the embodiment is specifically realized, and constructs the data identification model based on the data after feature extraction. The conventional unmanned aerial vehicle power data anomaly recognition technology generally adopts a traditional neural network model to train a classifier, however, the mode is easy to overfit, and the generalization capability of the model is reduced. In addition, the learning rate adjustment strategy in the prior art is single, which is not beneficial to rapid convergence of the model. Specifically, the invention provides a functional link neural network algorithm based on Riemann self-coding, which is used for an unmanned aerial vehicle power data anomaly identification task. Fig. 2 shows a flowchart of another method for identifying abnormal power data of an unmanned aerial vehicle based on a neural network, which is provided by the embodiment of the invention, as shown in fig. 2, and includes the following steps:
step S202, a preset training sample set is obtained.
Step S204 represents the training sample set as points on the Riemann manifold to encode the training sample set as Riemann manifold coding features.
In Riemann manifoldEach point has a local coordinate system, set with Riemann metric/>To measure the distance between two points on this manifold.
Further, input data is providedRepresented as a point on the Riemann manifold. By reference points/>And input data points/>Distance between Riemann/>To encode the input data. This distance can be calculated as:
Step S206, generating the self-attention vector corresponding to the Riemann manifold coding feature.
In order to improve model classification accuracy and enhance robustness and generalization capability of a model, the embodiment of the invention maps Riemann manifold coding features into a high-dimensional feature space and determines high-order feature vectors. And calculating the self-attention weight matrix according to the dimension of the high-order feature vector. And then, carrying out weighted aggregation on the high-order feature vectors based on the self-attention weight matrix to obtain the self-attention vector corresponding to the Riemann manifold coding feature.
In particular, embodiments of the present invention use a basis functionTo map the encoded features into a high-dimensional feature space, generating high-order features. The present invention uses polynomial basis functions, and this mapping can be expressed as: z= [1, E (x) 2,...,E(x)n ]. Where n is the order of the polynomial.
Further, in obtaining the high-order feature vectorAfter that, self-attention enhancement was performed. For eigenvectors/>The self-attention weight matrix/>, is calculated using the following manner:
Wherein,Is the eigenvector/>I.e. the number of features.
Further, a weight matrix is usedFor feature vector/>Performing weighted aggregation to obtain a self-attention vector z': z' =az.
Step S208, the self-attention vector is input into a preset function link neural network classifier, and the function link neural network classifier is trained.
Further, self-attention feature vectorInput into a functionally linked neural network classifier. The output of the functionally linked neural network can be written as:
wherein, Is/>Weights,/>Is biased,/>Is a Sigmoid activation function. The feature of the functionally linked neural network is that it combines the raw input data with a set of basis functions to generate high-order features as inputs.
Step S210, calculating a Laplace matrix of the self-attention vector, and calculating cross entropy loss of the classifier according to the Laplace matrix.
In an embodiment of the invention, a loss function is definedIs a cross entropy loss and is minimized by a preset optimization method, such as a gradient descent method, to train the model. Specifically, the loss function of the embodiment of the invention comprises Laplacian regularization term/>Its effect is to enhance the generalization ability of the model and prevent overfitting.
In Laplace regularization termIn/>Is a regularization parameter,/>Is the trace of the matrix,/>Is a feature matrix,/>Is a laplace matrix, and the calculation mode can be expressed as:
wherein, For the degree matrix, each element in the degree matrix/>Is/>The degree of each feature can be calculated by the following way:
wherein, Representing feature similarity,/>Is the number of features. The manner in which feature similarity is calculated can be expressed as:
wherein, And/>Is the feature matrix/>Two eigenvectors of >/>Is the width parameter of the gaussian kernel.
Further, the loss function of the embodiment of the inventionCan be expressed as:
wherein, Is the number of data points,/>Is/>True label of data points,/>Is a Laplacian regularization term.
Step S212, optimizing parameters of the classifier based on the cross entropy loss; and adaptively adjusting the learning rate of the classifier.
Step S214, until the classifier meets the preset training conditions, a data identification model is built based on the classifier.
Embodiments of the present invention minimize the loss function using gradient descent methodsWhen first initialize the weights/>And bias/>They are then updated in each iteration to reduce the value of the loss function.
Further, the update rules of the weights and biases are as follows:
wherein, Is learning rate,/>Is a loss function/>With respect to weights/>Is a gradient of (a). /(I)Is a loss function/>With respect to bias/>Is a gradient of (a).
In the invention, the learning rateIs dynamically adjusted in an adaptive manner. Specifically, the invention provides a cosine annealing-based method for dynamically adjusting the learning rate. Cosine annealing simulates a cosine cycle change in which the learning rate decreases from an initial value to a minimum value in one cycle and then resets in the next cycle. The cosine annealing formula is as follows:
wherein, Is at time step/>Learning rate of/>Is the minimum value of learning rate,/>Is the maximum value of the learning rate,Is the current time step,/>Is the length of one cycle.
At each training step, the gradients of the model parameters are calculated and updated:
wherein, Is a loss function,/>Is a model parameter.
Further, dynamically adjusted learning rate is usedUpdating parameters:
The cosine annealing formula described above is used to calculate the learning rate during each cycle of training And updating the model parameters according to the gradient.
By repeatedly and iteratively updating the weights and the offsets, a model capable of accurately identifying the abnormal power data of the unmanned aerial vehicle, namely a trained data identification model, can be obtained.
The embodiment of the invention provides another unmanned aerial vehicle power data abnormality recognition method based on a neural network, and provides a function link neural network algorithm based on Riemann self-coding, which is used for unmanned aerial vehicle power data abnormality recognition tasks. Riemann metrics are introduced to encode the input data and feature enhancement is performed in conjunction with a self-attention mechanism. In addition, a cosine annealing-based method for dynamically adjusting the learning rate is provided.
Furthermore, on the basis of the above embodiment, the embodiment of the present invention further provides a third method for identifying abnormal power data of an unmanned aerial vehicle based on a neural network, and the embodiment of the present invention describes the steps of constructing a training sample set. The data source unmanned aerial vehicle power system comprises data of a plurality of monitoring points, wherein each monitoring point has data of a plurality of attributes. Each attribute corresponds to a different data characteristic. It can be appreciated that in the task of identifying the abnormal power data of the unmanned aerial vehicle, the acquisition of the training samples is quite costly, and the identification accuracy is easily affected when the number of the training samples is insufficient. The embodiment of the invention also carries out data expansion on the acquired data so as to construct a training sample set.
However, the prior art mainly uses simple over-sampling and under-sampling techniques in terms of data expansion, which cannot generate data highly correlated with the original data, resulting in a model that is easily over-fitted or under-fitted. Since there is typically less abnormal data in the identification of anomalies in unmanned aerial vehicle power data, a more efficient data expansion method is needed to balance the data distribution. The invention provides an improved method of a self-adaptive synthetic sampling (ADASYN) algorithm based on quantum coding, which adopts a self-adaptive technology to carry out synthetic sampling on an original dataset, so that the synthetic data is more suitable for training of a neural network.
Fig. 3 shows a flowchart of a third method for identifying abnormal power data of an unmanned aerial vehicle based on a neural network, which is provided by an embodiment of the present invention, as shown in fig. 3, and includes the following steps:
Step S302, acquiring a pre-acquired power sample of the unmanned aerial vehicle.
And step S304, marking data of the unmanned aerial vehicle power sample according to the abnormal condition of the unmanned aerial vehicle power system corresponding to the unmanned aerial vehicle power sample, and generating a data tag.
In specific implementation, the unmanned aerial vehicle power sample refers to the unmanned aerial vehicle power data of the above embodiment, and will not be described here again. The method and the device for marking the collected data are characterized in that the collected data are marked as normal or abnormal according to historical abnormal conditions of the power data of the unmanned aerial vehicle. Each piece of data has a label L, wherein 0 represents normal data and 1 represents abnormal data.
In a specific embodiment, the monitoring point has 10 attributes: voltage (V), current (a), temperature (c), humidity (%), wind speed (m/s), wind direction (degree), solar radiation intensity (W/m < m >), power Factor (PF), active power (kW), reactive power (kVAR), respectively denoted p 1,p2,…,p10.
For a certain timestamp t, there is a data tuple d= (t, p 1,p2,…,p10).
Let the data of the monitoring point be :p1=220V;p2=5A;p3=25℃;p4=45%;p5=2.5m/s;p6=180°;p7=600W/m²;p8=0.95PF;p9=15kW;p10=8kVAR; and the label be l=0 at time t, which means that this is a normal data. Then, for this data, the data of the monitoring point can be used as a feature vector, namely:=[220,5,25,45,2.5,180,600,0.95,15,8]。
the tag of this data acts as an output vector, namely: =[0]。
And step S306, vectorizing the unmanned aerial vehicle power sample and the data tag to construct an initial sample set.
Each data tupleCan be expressed as a feature vector/>Tag vector/>. Then the entire dataset can be represented as a matrix/>And a vector/>Wherein/>Is a feature vector of data corresponding to each row of the data,/>A tag for one data for each element of (a) can be expressed as:
wherein, Is the number of data bars in the dataset,/>Is/>First/>, of stripe dataNumerical value of the attribute,/>Is/>Tag of the stripe data.
Further, it can be appreciated that in tasks of abnormality recognition of unmanned aerial vehicle power data, problems of non-normative, abnormal values and the like may exist in the data. In order to improve the data quality and ensure the training effect of the model, the invention preprocesses the acquired data.
Firstly, the data is normalized, and in the process of data normalization, the first data set is setThe maximum value of the individual features is/>Minimum value is/>. Will/>Each data point of each feature/>Normalization is as follows:
wherein, Is a manually preset hyper-parameter. /(I)For characteristic values before normalization,/>Is the normalized characteristic value.
Further, data outlier processing is performed. The invention provides a method based on local anomaly factors. For data setsEach data point/>And calculating local abnormality factors of the local abnormality factors.
Specifically, first calculateDistance from other data points, and choose/>The nearest neighbor data points are recorded as. Calculation/>Can reach the distance of (3):
wherein, Is/>To/>Distance of neighbor,/>Is/>To/>The distance of any data point in the database.
Further, calculateCan be expressed as:
Further, a local anomaly factor is calculated:
Data points with larger local anomaly factors are regarded as anomaly values and are rejected.
For preprocessed data setsWherein/>Is a data point,/>Is a label of the data point. Further, the data is augmented in the following manner to construct a training sample set.
In step S308, the initial sample set is converted into a quantum form by using quantum encoding technology, so as to obtain quantum encoded data.
When the embodiment of the invention expands data, firstly, the original data point is converted into a quantum form by using a quantum coding technology. And then using a data expansion algorithm to expand the data.
The quantum coding is a technology for converting data points into quantum states, and the basic idea is to code each characteristic value of the data points onto one quantum bit, and realize information transmission and processing through the characteristics of a quantum system.
Specifically, for data pointsThe quantum coding form is as follows:
wherein, Is the dimension of the data point,/>Is the data point/>(1 /)Characteristic value/>Is the tensor product.
Step S310, carrying out data expansion on the quantum coding data through a preset data expansion algorithm, and constructing a training sample set.
In particular implementations, embodiments of the present invention use an improved adaptive synthesis sampling algorithm to synthesize samples of quantum encoded data to generate a first set of extended samples. Further, data expansion is carried out on the quantum coded data by applying preset quantum gate operation, and a second expansion sample set is generated. And combining the first extended sample set, the second extended sample set and the initial sample set to obtain a training sample set.
1) A first set of extended samples is obtained based on the quantum encoded data. The embodiment of the invention uses the improved ADASYN (Adaptive Synthetic Sampling) algorithm to synthesize and sample the data after quantum encoding. When the method is specifically implemented, firstly, the classification difficulty coefficient corresponding to the first class data in the quantum coding data is calculated; and then, calculating the number of the synthesized data corresponding to the first class data based on the classification difficulty coefficient. Further, based on the number of the synthesized data and the neighbor samples corresponding to the first class data, data synthesis is performed to obtain synthesized data points. And taking the synthesized data points with similarity meeting the preset threshold as a first expansion sample to construct a first expansion sample set.
Specifically, the difficulty coefficient of each minority class data point is calculated firstly:
Wherein,Is the data point/>/>Neighbor,/>Is the data point/>Is a label of (a). /(I)The larger the value of (2), the description data pointThe harder it is to classify.
Further, the composite data quantity of each minority class data point is calculated:
Wherein,Is rounded up,/>Is the total composite data amount.
Further, for each minority class data point, use itThe neighbor data points generate composite data. Specifically, for data points/>Selecting a random neighbor data point/>Generate synthetic data points/>:
Wherein,Is a randomly selected interpolation coefficient.
Further, the inner product of the quantum states is used to calculate the similarity of two data points:
wherein, And/>Data points/>, respectivelyAnd/>(1 /)And characteristic values. The inner product of the quantum states is used to better reflect the similarity of the data points.
Further, judgeAnd/>If the similarity of the (E) is satisfied with a preset threshold, reserving a newly generated sample/>, if soThe first expansion sample is obtained after multiple expansion; otherwise, discard the newly generated samples/>。
2) Further, a second set of extended samples is obtained based on the quantum encoded data. Specifically, the embodiment of the invention uses a Hadamard gate to expand data based on the data after quantum encoding. Quantum gates are a basic unit for manipulating qubits in quantum computing. Common quantum gates include Pauli-X gates, pauli-Y gates, pauli-Z gates, hadamard gates, and the like. The quantum gate can realize the linear transformation of the quantum state and operate the quantum state.
The embodiment of the invention firstly applies Hadamard quantum gate operation to quantum coding data, expands the quantum state of the quantum coding data and obtains quantum state expansion data; further, format conversion is carried out on the quantum state expansion data through measurement operation, so that classical data are obtained; thereafter, a second set of expanded samples is constructed based on the classical data.
The invention uses Hadamard gates for data expansion. The matrix form of Hadamard gates is:
The original quantum encoded data is extended to more quantum states using Hadamard gates.
Specifically, quantum coded data that needs to be expanded is first selected.
Let the selected data point beThe quantum coding form is as follows:
further, hadamard gate operation is applied.
For each qubit of the data point, a Hadamard gate operation is applied, which can be expressed as:
further, after the Hadamard gate is applied, the original quantum state is expanded into more quantum states. Specifically, for each qubit of the original quantum state, two new quantum states can be obtained. Thus, the expanded number of original quantum states will be Wherein/>Is the dimension of the data point.
Further, the extended quantum states are converted back into classical data.
For each extended quantum state, it is converted back into classical data by a measurement operation. In particular, for extended quantum statesClassical data/>, were obtained by the following measurement procedureAs the second expansion data, it can be expressed as:
further, second augmentation data is added to the dataset.
Through the above steps, an expanded data set can be obtainedWherein/>Is the number of extended data,/>Is the amount of raw data.
3) In summary, the first extended sample set, the second extended sample set, and the initial sample set are combined to obtain an extended data set, i.e., a training sample set, which includes the original data and the synthesized data. Meanwhile, the quantum coding can preserve the high-order characteristics of the data points, and the improved ADASYN algorithm can generate the synthesized data according to the difficulty coefficient of the data points, so that the synthesized data are more uniformly distributed in the space of a few data points.
The third method for identifying abnormal unmanned aerial vehicle power data based on the neural network provided by the embodiment of the invention provides an improved method of a self-adaptive synthetic sampling (ADASYN) algorithm based on quantum coding. The raw data set is synthetically sampled using adaptive techniques, making the synthetic data more suitable for training of neural networks. The data expansion is carried out by adopting quantum coding and quantum gate, so that the expanded data has more diversity, and the high-order characteristics of the data points are reserved.
Example III
Furthermore, on the basis of the above embodiment, the embodiment of the present invention further provides a fourth method for identifying abnormal power data of an unmanned aerial vehicle based on a neural network, which describes a method for constructing a feature extraction model. According to the embodiment, the characteristic extraction model is used for extracting the characteristics of the power data of the unmanned aerial vehicle, and the key characteristics are determined. Traditional feature extraction methods typically rely on artificial feature engineering, are time-consuming and labor-consuming, and are susceptible to human subjective factors. In addition, the conventional neural network model is usually optimized by using a gradient descent algorithm, so that the model is easy to fall into a locally optimal solution, and the generalization capability of the model is poor. The invention provides a neural network algorithm based on harmonic resonance optimization, which is used for avoiding the situation that a traditional gradient descent algorithm is easy to fall into a local optimal solution in the training process of the neural network.
In a specific implementation, fig. 4 shows a flowchart of a fourth method for identifying abnormal power data of an unmanned aerial vehicle based on a neural network, where the method includes the following steps as shown in fig. 4:
step S402, a preset training sample set is obtained.
Step S404, training a preset neural network through a training sample set.
Step S406, determining a weighted second derivative of the neural network, and calculating a loss function of the neural network based on the weighted second derivative.
In specific implementation, the neural network may be trained by the training sample set constructed in the above embodiments. The feature extraction model of the invention is a neural network, which consists of a plurality of neurons, and each neuron has corresponding weight and bias. The weights and biases are updated by an optimization algorithm. Let the weight matrix of the neural network beBias matrix is/>. In the training process of the neural network, a loss function is set as/>Wherein/>And/>Respectively weight and bias.
Specifically, the loss function may be expressed as:
wherein, Is a cross entropy loss function,/>Is a constant preset by personnel,/>Is the number of weights. Loss function/>The second derivative of the weight is included, so that the change of the weight can be reflected better, and the optimization effect can be improved.
Step S408, optimizing network parameters of the neural network through a preset harmonic resonance algorithm, and adaptively adjusting the learning rate of the neural network according to the harmonic resonance coefficient corresponding to the harmonic resonance algorithm. Wherein the neural network optimizes network parameters based on the fade harmonic activation function.
When the embodiment of the invention performs model training, on one hand, the network parameters of the neural network are optimized based on the harmonic resonance algorithm, and the method is inspired by the harmonic resonance theory, namely, the harmonic resonance phenomenon is that when the frequency of an external signal is matched with the natural frequency of a system, the amplitude of the system is increased sharply. The present invention uses this principle to optimize the parameters of the neural network.
On the other hand, the learning rate is adaptively adjusted according to the harmonic resonance coefficient corresponding to the harmonic resonance algorithm. And, network parameters are optimized based on the fade harmonic activation function.
1) Network parameters of the neural network are optimized based on a harmonic resonance algorithm:
In specific implementation, in the training process of the neural network, the embodiment of the invention regards the change of the loss function as external driving force, regards the change of the network parameter as natural frequency, and calculates the external driving force and the harmonic resonance coefficient corresponding to the natural frequency.
In a specific embodiment, consider a simple vibrator whose equation of motion is:
wherein, Is the position of the vibrator,/>Is the natural frequency/>Is an external driving force. Then this equation can be rewritten as:
wherein, Is an external periodic driving force.
The embodiment of the invention introduces a new optimization algorithm which takes the external driving force as a loss functionThe natural frequency is considered as a change in weight and bias. A threshold will be set and when the amplitude reaches this threshold, the optimization is stopped. Specifically, a new harmonic resonance coefficient/>, is defined:
Wherein,Is the second derivative of the loss function,/>Is the optimized frequency,/>Is an optimization cycle.
Setting the optimized frequency to a frequency matching the weight and bias variation and adjusting step by step untilThe threshold is reached.
Further, when the harmonic resonance coefficient reaches a preset optimization threshold, updating the harmonic resonance coefficient based on the loss function so as to enable the neural network to reach a preset training condition. Specifically, after R reaches the preset threshold (i.e., the optimization threshold), the following formula is used to update the harmonic resonance coefficient in each iterative optimization cycle according to the embodiment of the present invention:
Wherein,Is/>Harmonic resonance coefficient of each optimization period,/>And/>Is a constant,/>Is the first derivative of the loss function.
2) According to the harmonic resonance coefficient corresponding to the harmonic resonance algorithm, the learning rate of the neural network is adaptively adjusted: wherein in each optimization cycle, the resonance coefficients will be tuned according toAdaptively adjusting learning rate/>Learning rate/>The update rule of (2) is as follows: /(I)
Wherein,Is/>Learning rate of each optimization cycle; /(I)Is a constant for updating the learning rate and is preset by human beings.
3) The neural network optimizes network parameters based on the gradual-change harmonic activation function:
In specific implementation, the embodiment of the invention defines a basic harmonic wave function according to the neuron value of the neural network, and defines a harmonic amplitude function according to the absolute value of the neuron value of the neural network; multiplying the harmonic amplitude function with the basic harmonic wave function to define a gradual change harmonic activation function; calculating the gradient of the gradual change and activation functions; the network parameters are updated based on the gradient of the fade and activation functions.
In particular, in training the neural network, the purpose of the training is to optimize the weights of the neural network in a continuous iterative processAnd bias/>. For neurons in a neural network, each neuron functions by virtue of its activation function, a gradual change and activation function is proposed in the present invention.
In the neural network of the present invention, the activation function is used to introduce nonlinearities so that the neural network can approach complex functions.
First, define a basic harmonic wave function:
Wherein,Is the neuron value of the input,/>Is a phase shift parameter.
Then, a harmonic amplitude function is defined:
Wherein,Is the absolute value of the input neuron value.
Further, multiplying the harmonic amplitude function with the base harmonic wave function to define a gradual harmonic activation function:
Further, in order to use this new activation function in a neural network, its gradient needs to be calculated. The present invention uses the chain law to calculateIs a gradient of (a).
Specifically, first calculateIs the derivative of: /(I)
Further, calculateIs the derivative of: /(I)
Further, calculation using the chain lawIs a gradient of (2):
further, the weights and offsets are updated. Updating weights using gradient of gradual change harmonic activation function And bias/>The update style can be expressed as:
/>
Step S410, when the harmonic resonance coefficient reaches a preset training threshold, the number of hidden layers of the neural network is adjusted.
The neural network structure of the embodiment of the invention is determined through self-adaption, and the invention introduces a mechanism for adjusting the self-adaption neural network structure. When the resonance coefficient is adjustedWhen a predetermined threshold (i.e., a training threshold) is reached, the hidden layer of the neural network will automatically be increased or decreased. Specifically, the following formula is used to decide to increase or decrease the number of hidden layers:
wherein, Is the amount of change in the number of hidden layers,/>Is a constant,/>Is the threshold for the harmonic resonance coefficient.
And step S412, until the neural network reaches a preset training condition, constructing a feature extraction model based on the neural network.
The fourth unmanned aerial vehicle power data anomaly identification method based on the neural network provided by the embodiment of the invention provides a neural network algorithm based on harmonic resonance optimization, introduces a self-adaptive neural network structure adjustment mechanism, and provides a gradual change harmonic activation function, wherein the harmonic resonance theory is utilized to optimize the neural network parameters, so that the condition that the traditional gradient descent algorithm is easy to fall into local optimal solution can be avoided.
In summary, the embodiment of the invention adopts quantum coding and quantum gate to expand data, so that the synthesized data is more uniformly distributed in the space of few data points, thereby improving the robustness of the model. Meanwhile, the neural network algorithm based on harmonic resonance optimization improves the optimization effect of the model and improves the generalization capability of the model. Therefore, the embodiment of the invention can better identify the abnormality of the power data of the unmanned aerial vehicle, and improves the identification precision.
Example IV
Further, on the basis of the above method embodiment, the embodiment of the present invention further provides a device for identifying abnormal power data of an unmanned aerial vehicle based on a neural network, and fig. 5 shows a schematic structural diagram of the device for identifying abnormal power data of an unmanned aerial vehicle based on a neural network provided by the embodiment of the present invention, as shown in fig. 5, where the device includes: the data acquisition module 100 is used for acquiring power data of the unmanned aerial vehicle to be identified; the data processing module 200 is used for extracting characteristics of the power data of the unmanned aerial vehicle and determining key characteristics in the power data of the unmanned aerial vehicle; the execution module 300 is used for identifying key features through a pre-constructed data identification model and determining an identification result corresponding to the unmanned aerial vehicle power data; the training sample set of the training data recognition model comprises a training sample and a sample label, wherein the sample label is used for representing the abnormal condition corresponding to the training sample; the training sample is constructed based on data of a plurality of monitoring points of the unmanned aerial vehicle power system, and data expansion is carried out by adopting quantum coding and quantum gates; a data conversion module 400, configured to convert the recognition result into a confidence score; and the output module 500 is used for determining the abnormal situation corresponding to the unmanned aerial vehicle power data based on the recognition result when the confidence score meets the preset confidence threshold.
The unmanned aerial vehicle power data abnormality recognition device based on the neural network provided by the embodiment of the invention has the same technical characteristics as the method embodiment, so that the same technical problems can be solved, and the same technical effects can be achieved.
Further, based on the foregoing embodiment, the embodiment of the present invention further provides another device for identifying abnormal power data of an unmanned aerial vehicle based on a neural network, and fig. 6 shows a schematic structural diagram of another device for identifying abnormal power data of an unmanned aerial vehicle based on a neural network provided by the embodiment of the present invention, and referring to fig. 6, the execution module 300 is further configured to obtain a preset training sample set; representing the training sample set as points on the Riemann manifold to encode the training sample set as Riemann manifold coding features; generating a self-attention vector corresponding to the Riemann manifold coding feature; inputting the self-attention vector into a preset functional link neural network classifier, and training the functional link neural network classifier; calculating a Laplace matrix of the self-attention vector, and calculating cross entropy loss of the classifier according to the Laplace matrix; optimizing parameters of the classifier based on the cross entropy loss; and adaptively adjusting the learning rate of the classifier; until the classifier meets preset training conditions, a data identification model is built based on the classifier.
The execution module 300 is further configured to map the risman manifold coding feature into a high-dimensional feature space, and determine a high-order feature vector; calculating a self-attention weight matrix according to the dimension of the high-order feature vector; and carrying out weighted aggregation on the high-order feature vectors based on the self-attention weight matrix to obtain the self-attention vectors corresponding to the Riemann manifold coding features.
Further, the embodiment of the invention performs feature extraction on the unmanned aerial vehicle power data through a pre-constructed feature extraction model; the data processing module 200 is further configured to obtain a preset training sample set; training a preset neural network through a training sample set; determining a weighted second derivative of the neural network, and calculating a loss function of the neural network based on the weighted second derivative; optimizing network parameters of the neural network through a preset harmonic resonance algorithm, and adaptively adjusting the learning rate of the neural network according to harmonic resonance coefficients corresponding to the harmonic resonance algorithm; the neural network optimizes network parameters based on the gradual change harmonic activation function; when the harmonic resonance coefficient reaches a preset training threshold, adjusting the number of hidden layers of the neural network; and constructing a feature extraction model based on the neural network until the neural network reaches a preset training condition.
The data processing module 200 is further configured to, in a training process of the neural network, consider a change of the loss function as an external driving force, and consider a change of a network parameter as a natural frequency, and calculate a harmonic resonance coefficient corresponding to the external driving force and the natural frequency; when the harmonic resonance coefficient reaches a preset optimization threshold, updating the harmonic resonance coefficient based on the loss function so as to enable the neural network to reach a preset training condition.
The data processing module 200 is further configured to define a basic harmonic wave function according to a neuron value of the neural network, and define a harmonic amplitude function according to an absolute value of the neuron value of the neural network; multiplying the harmonic amplitude function with the basic harmonic wave function to define a gradual change harmonic activation function; calculating the gradient of the gradual change and activation functions; the network parameters are updated based on the gradient of the fade and activation functions.
Further, the device further comprises a construction module 600, configured to obtain a pre-collected power sample of the unmanned aerial vehicle; the unmanned aerial vehicle power sample comprises data for collecting a plurality of monitoring points of the unmanned aerial vehicle power system; according to the abnormal condition of the unmanned aerial vehicle power system corresponding to the unmanned aerial vehicle power sample, carrying out data marking on the unmanned aerial vehicle power sample to generate a data tag; vectorizing an unmanned aerial vehicle power sample and a data tag to construct an initial sample set; converting the initial sample set into a quantum form by using a quantum coding technology to obtain quantum coding data; and carrying out data expansion on the quantum coding data through a preset data expansion algorithm to construct a training sample set.
The above-mentioned construction module 600 is further configured to synthetically sample the quantum encoded data using a modified adaptive synthetic sampling algorithm to generate a first extended sample set; performing data expansion on the quantum coded data by applying preset quantum gate operation to generate a second expansion sample set; and combining the first extended sample set, the second extended sample set and the initial sample set to obtain a training sample set.
The above construction module 600 is further configured to calculate a classification difficulty coefficient corresponding to the first type of data in the quantum encoded data; based on the classification difficulty coefficient, calculating the number of synthesized data corresponding to the first class data; based on the number of the synthesized data and the neighbor samples corresponding to the first class data, carrying out data synthesis to obtain synthesized data points; and taking the synthesized data points with the similarity meeting the preset threshold as a first expansion sample to construct a first expansion sample set.
The above construction module 600 is further configured to perform data expansion on the quantum encoded data by applying a preset quantum gate operation, and generate a second expansion sample set, where the step includes: applying Hadamard quantum gate operation to the quantum coded data, and expanding the quantum state of the quantum coded data to obtain quantum state expansion data; carrying out format conversion on the quantum state expansion data through measurement operation to obtain classical data; a second extended sample set is constructed based on the classical data.
The embodiment of the invention also provides an electronic device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to realize the steps of the method shown in the figures 1 to 4. The embodiments of the present invention also provide a computer readable storage medium having a computer program stored thereon, which when executed by a processor performs the steps of the method shown in fig. 1 to 4 described above.
The embodiment of the present invention further provides a schematic structural diagram of an electronic device, as shown in fig. 7, where the electronic device includes a processor 71 and a memory 70, where the memory 70 stores computer executable instructions that can be executed by the processor 71, and the processor 71 executes the computer executable instructions to implement the methods shown in fig. 1 to 4.
In the embodiment shown in fig. 7, the electronic device further comprises a bus 72 and a communication interface 73, wherein the processor 71, the communication interface 73 and the memory 70 are connected by the bus 72. The memory 70 may include a high-speed random access memory (RAM, random Access Memory), and may further include a non-volatile memory (non-volatile memory), such as at least one magnetic disk memory. The communication connection between the system network element and the at least one other network element is achieved via at least one communication interface 73 (which may be wired or wireless), which may use the internet, a wide area network, a local network, a metropolitan area network, etc. The Bus 72 may be an ISA (Industry Standard Architecture ) Bus, a PCI (PERIPHERAL COMPONENT INTERCONNECT, peripheral component interconnect standard) Bus, or EISA (Extended Industry Standard Architecture ) Bus, etc., or an AMBA (Advanced Microcontroller Bus Architecture, standard for on-chip buses) Bus, where AMBA defines three buses, including an APB (ADVANCED PERIPHERAL Bus) Bus, an AHB (ADVANCED HIGH-performance Bus) Bus, and a AXI (Advanced eXtensible Interface) Bus. The bus 72 may be classified as an address bus, a data bus, a control bus, or the like. For ease of illustration, only one bi-directional arrow is shown in FIG. 7, but not only one bus or type of bus.
The processor 71 may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuitry in hardware or instructions in software in the processor 71. The processor 71 may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU for short), a network processor (Network Processor, NP for short), and the like; but may also be a digital signal Processor (DIGITAL SIGNAL Processor, DSP), application Specific Integrated Circuit (ASIC), field-Programmable gate array (FPGA) or other Programmable logic device, discrete gate or transistor logic device, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present application may be embodied directly in the execution of a hardware decoding processor, or in the execution of a combination of hardware and software modules in a decoding processor. The software modules may be located in a random access memory, flash memory, read only memory, programmable read only memory, or electrically erasable programmable memory, registers, etc. as well known in the art. The storage medium is located in a memory and the processor 71 reads the information in the memory and in combination with its hardware performs the method as shown in any of the foregoing figures 1 to 4.
The computer program product of the method and the device for identifying abnormal power data of an unmanned aerial vehicle based on the neural network provided by the embodiment of the invention comprises a computer readable storage medium storing program codes, wherein the instructions included in the program codes can be used for executing the method described in the method embodiment, and specific implementation can be seen in the method embodiment and is not repeated here. Finally, it should be noted that: the above examples are only specific embodiments of the present invention for illustrating the technical solution of the present invention, but not for limiting the scope of the present invention, and although the present invention has been described in detail with reference to the foregoing examples, it will be understood by those skilled in the art that the present invention is not limited thereto: any person skilled in the art may modify or easily conceive of the technical solution described in the foregoing embodiments, or perform equivalent substitution of some of the technical features, while remaining within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention, and are intended to be included in the scope of the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.
Claims (10)
1. The unmanned aerial vehicle power data anomaly identification method based on the neural network is characterized by comprising the following steps of:
acquiring power data of an unmanned aerial vehicle to be identified;
extracting features of the unmanned aerial vehicle power data, and determining key features in the unmanned aerial vehicle power data;
Identifying the key features through a pre-constructed data identification model, and determining an identification result corresponding to the unmanned aerial vehicle power data; the training sample set for training the data recognition model comprises a training sample and a sample label, wherein the sample label is used for representing an abnormal condition corresponding to the training sample; the training sample is constructed based on data of a plurality of monitoring points of an unmanned aerial vehicle power system, and data expansion is carried out by adopting quantum coding and quantum gates;
Converting the identification result into a confidence score;
and when the confidence score meets a preset confidence threshold, determining an abnormal condition corresponding to the unmanned aerial vehicle power data based on the identification result.
2. The method according to claim 1, wherein the method for constructing the data identification model comprises:
acquiring a preset training sample set;
Representing the training sample set as points on a Riemann manifold to encode the training sample set as Riemann manifold coding features;
Generating a self-attention vector corresponding to the Riemann manifold coding feature;
inputting the self-attention vector into a preset functional link neural network classifier, and training the functional link neural network classifier;
Calculating a Laplace matrix of the self-attention vector, and calculating cross entropy loss of the classifier according to the Laplace matrix;
optimizing parameters of the classifier based on the cross entropy loss; and adaptively adjusting the learning rate of the classifier;
until the classifier meets preset training conditions, a data identification model is built based on the classifier.
3. The method of claim 2, wherein the step of generating the self-attention vector corresponding to the risman manifold-coded feature comprises:
mapping the Riemann manifold coding feature into a high-dimensional feature space, and determining a high-order feature vector;
calculating a self-attention weight matrix according to the dimension of the high-order feature vector;
and carrying out weighted aggregation on the high-order feature vectors based on the self-attention weight matrix to obtain self-attention vectors corresponding to the Riemann manifold coding features.
4. The method according to claim 1, wherein the unmanned aerial vehicle power data is feature extracted by a pre-constructed feature extraction model; the construction method of the feature extraction model comprises the following steps:
acquiring a preset training sample set;
training a preset neural network through the training sample set;
determining a weighted second derivative of the neural network, and calculating a loss function of the neural network based on the weighted second derivative;
optimizing network parameters of the neural network through a preset harmonic resonance algorithm, and adaptively adjusting the learning rate of the neural network according to harmonic resonance coefficients corresponding to the harmonic resonance algorithm; wherein the neural network optimizes the network parameters based on a gradual change harmonic activation function;
When the harmonic resonance coefficient reaches a preset training threshold, adjusting the number of hidden layers of the neural network;
And constructing a feature extraction model based on the neural network until the neural network reaches a preset training condition.
5. The method of claim 4, wherein optimizing the network parameters of the neural network by a preset harmonic resonance algorithm comprises:
In the training process of the neural network, taking the change of the loss function as an external driving force, taking the change of the network parameter as a natural frequency, and calculating a harmonic resonance coefficient corresponding to the external driving force and the natural frequency;
And updating the harmonic resonance coefficient based on a loss function when the harmonic resonance coefficient reaches a preset optimization threshold value, so that the neural network reaches a preset training condition.
6. The method of claim 4, wherein the step of optimizing the network parameters by the neural network based on a fade harmonic activation function comprises:
defining a basic harmonic wave function according to the neuron value of the neural network, and defining a harmonic amplitude function according to the absolute value of the neuron value of the neural network;
multiplying the harmonic amplitude function with the basic harmonic wave function to define a gradual change harmonic activation function;
Calculating the gradient of the gradual change and activation function;
Updating the network parameters based on the gradient of the fade and activation functions.
7. The method according to claim 1, wherein the method further comprises:
Acquiring a pre-acquired unmanned aerial vehicle power sample; the unmanned aerial vehicle power sample comprises data for collecting a plurality of monitoring points of an unmanned aerial vehicle power system;
According to the abnormal condition of the unmanned aerial vehicle power system corresponding to the unmanned aerial vehicle power sample, carrying out data marking on the unmanned aerial vehicle power sample to generate a data tag;
Vectorizing the unmanned aerial vehicle power sample and the data tag to construct an initial sample set;
converting the initial sample set into a quantum form by using a quantum coding technology to obtain quantum coding data;
And carrying out data expansion on the quantum coding data through a preset data expansion algorithm to construct a training sample set.
8. The method of claim 7, wherein the step of data expanding the quantum encoded data by a predetermined data expansion algorithm to construct a training sample set comprises:
Synthesizing and sampling the quantum coded data by using an improved self-adaptive synthesis and sampling algorithm to generate a first expansion sample set;
performing data expansion on the quantum coded data by applying a preset quantum gate operation to generate a second expansion sample set;
and combining the first extended sample set, the second extended sample set and the initial sample set to obtain a training sample set.
9. The method of claim 8, wherein the step of synthesizing the quantum encoded data using a modified adaptive synthesis sampling algorithm to generate a first set of expanded samples comprises:
calculating a classification difficulty coefficient corresponding to first class data in the quantum coding data;
based on the classification difficulty coefficient, calculating the amount of synthesized data corresponding to the first class data;
based on the synthesized data quantity and the neighbor samples corresponding to the first class data, carrying out data synthesis to obtain synthesized data points;
Taking the synthesized data points with the similarity meeting a preset threshold as a first expansion sample to construct a first expansion sample set;
performing data expansion on the quantum coded data by applying a preset quantum gate operation, and generating a second expansion sample set, wherein the method comprises the following steps:
applying Hadamard quantum gate operation to the quantum coding data, and expanding the quantum state of the quantum coding data to obtain quantum state expansion data;
performing format conversion on the quantum state expansion data through measurement operation to obtain classical data;
a second set of expanded samples is constructed based on the classical data.
10. An abnormal recognition device of unmanned aerial vehicle power data based on neural network, characterized in that, the device includes:
the data acquisition module is used for acquiring power data of the unmanned aerial vehicle to be identified;
The data processing module is used for extracting the characteristics of the unmanned aerial vehicle power data and determining key characteristics in the unmanned aerial vehicle power data;
The execution module is used for identifying the key features through a pre-constructed data identification model and determining an identification result corresponding to the power data of the unmanned aerial vehicle; the training sample set for training the data recognition model comprises a training sample and a sample label, wherein the sample label is used for representing an abnormal condition corresponding to the training sample; the training sample is constructed based on data of a plurality of monitoring points of an unmanned aerial vehicle power system, and data expansion is carried out by adopting quantum coding and quantum gates;
the data conversion module is used for converting the identification result into a confidence score;
and the output module is used for determining the abnormal situation corresponding to the unmanned aerial vehicle power data based on the identification result when the confidence score meets a preset confidence threshold.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410487195.0A CN118070204B (en) | 2024-04-23 | 2024-04-23 | Unmanned aerial vehicle power data anomaly identification method and device based on neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410487195.0A CN118070204B (en) | 2024-04-23 | 2024-04-23 | Unmanned aerial vehicle power data anomaly identification method and device based on neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN118070204A true CN118070204A (en) | 2024-05-24 |
CN118070204B CN118070204B (en) | 2024-07-19 |
Family
ID=91102423
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410487195.0A Active CN118070204B (en) | 2024-04-23 | 2024-04-23 | Unmanned aerial vehicle power data anomaly identification method and device based on neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN118070204B (en) |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020143321A1 (en) * | 2019-01-09 | 2020-07-16 | 平安科技(深圳)有限公司 | Training sample data augmentation method based on variational autoencoder, storage medium and computer device |
WO2021205066A1 (en) * | 2020-04-09 | 2021-10-14 | Nokia Technologies Oy | Training a data coding system for use with machines |
WO2022036801A1 (en) * | 2020-08-18 | 2022-02-24 | 深圳大学 | Method and system for achieving coexistence of heterogeneous networks |
CN115130536A (en) * | 2022-04-08 | 2022-09-30 | 腾讯科技(深圳)有限公司 | Training method of feature extraction model, data processing method, device and equipment |
CN115457414A (en) * | 2022-09-15 | 2022-12-09 | 西华大学 | Unmanned aerial vehicle abnormal behavior identification method based on improved residual error network |
CN115909095A (en) * | 2022-10-31 | 2023-04-04 | 广东电网有限责任公司广州供电局 | Color-based power transmission line key wire clamp defect identification method, system and medium |
CN116595463A (en) * | 2023-07-18 | 2023-08-15 | 国网山东省电力公司武城县供电公司 | Construction method of electricity larceny identification model, and electricity larceny behavior identification method and device |
CN116977834A (en) * | 2023-08-25 | 2023-10-31 | 杭州智元研究院有限公司 | Method for identifying internal and external images distributed under open condition |
CN117216644A (en) * | 2023-11-09 | 2023-12-12 | 北京世纪慈海科技有限公司 | Human body posture recognition method and device based on electric digital data processing |
-
2024
- 2024-04-23 CN CN202410487195.0A patent/CN118070204B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020143321A1 (en) * | 2019-01-09 | 2020-07-16 | 平安科技(深圳)有限公司 | Training sample data augmentation method based on variational autoencoder, storage medium and computer device |
WO2021205066A1 (en) * | 2020-04-09 | 2021-10-14 | Nokia Technologies Oy | Training a data coding system for use with machines |
WO2022036801A1 (en) * | 2020-08-18 | 2022-02-24 | 深圳大学 | Method and system for achieving coexistence of heterogeneous networks |
CN115130536A (en) * | 2022-04-08 | 2022-09-30 | 腾讯科技(深圳)有限公司 | Training method of feature extraction model, data processing method, device and equipment |
CN115457414A (en) * | 2022-09-15 | 2022-12-09 | 西华大学 | Unmanned aerial vehicle abnormal behavior identification method based on improved residual error network |
CN115909095A (en) * | 2022-10-31 | 2023-04-04 | 广东电网有限责任公司广州供电局 | Color-based power transmission line key wire clamp defect identification method, system and medium |
CN116595463A (en) * | 2023-07-18 | 2023-08-15 | 国网山东省电力公司武城县供电公司 | Construction method of electricity larceny identification model, and electricity larceny behavior identification method and device |
CN116977834A (en) * | 2023-08-25 | 2023-10-31 | 杭州智元研究院有限公司 | Method for identifying internal and external images distributed under open condition |
CN117216644A (en) * | 2023-11-09 | 2023-12-12 | 北京世纪慈海科技有限公司 | Human body posture recognition method and device based on electric digital data processing |
Non-Patent Citations (3)
Title |
---|
孙世宇;李喆;李建增;胡永江;: "基于卷积神经网络的无人机图像模糊类型识别", 火力与指挥控制, no. 02, 15 February 2020 (2020-02-15) * |
张鹏;: "基于运动解耦的无人机驱动故障自动检测***", 自动化与仪器仪表, no. 02, 25 February 2020 (2020-02-25) * |
贺思捷;刘大同;彭宇;: "遥测数据驱动的无人机飞行状态识别方法", 仪器仪表学报, no. 09, 15 September 2016 (2016-09-15) * |
Also Published As
Publication number | Publication date |
---|---|
CN118070204B (en) | 2024-07-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | A graph-cnn for 3d point cloud classification | |
CN113159239B (en) | Method for processing graph data by quantum graph convolutional neural network | |
Jia et al. | Feature dimensionality reduction: a review | |
CN110263227B (en) | Group partner discovery method and system based on graph neural network | |
CN109242223B (en) | Quantum support vector machine evaluation and prediction method for urban public building fire risk | |
CN107578061A (en) | Based on the imbalanced data classification issue method for minimizing loss study | |
CN108108854A (en) | City road network link prediction method, system and storage medium | |
Kong et al. | Hyperspectral image clustering based on unsupervised broad learning | |
CN115545350B (en) | Vehicle path problem solving method integrating deep neural network and reinforcement learning | |
CN117851921B (en) | Equipment life prediction method and device based on transfer learning | |
CN117251754A (en) | CNN-GRU energy consumption prediction method considering dynamic time packaging | |
Krishna et al. | Fuzzy-twin proximal SVM kernel-based deep learning neural network model for hyperspectral image classification | |
CN113541834A (en) | Abnormal signal semi-supervised classification method and system and data processing terminal | |
US20040078351A1 (en) | Non-linear data mapping and dimensionality reduction system | |
CN112199884A (en) | Article molecule generation method, device, equipment and storage medium | |
CN116975782A (en) | Hierarchical time sequence prediction method and system based on multi-level information fusion | |
An | Image classification algorithm based on stacked sparse coding deep learning model-optimized kernel function nonnegative sparse representation | |
CN118032327A (en) | Equipment intelligent lubrication monitoring method and device based on artificial intelligence | |
CN117407781A (en) | Equipment fault diagnosis method and device based on federal learning | |
CN118070204B (en) | Unmanned aerial vehicle power data anomaly identification method and device based on neural network | |
Zhang et al. | Semantic segmentation of spectral LiDAR point clouds based on neural architecture search | |
CN117290732A (en) | Construction method of fault classification model, wind power equipment fault classification method and device | |
Utku et al. | An efficient hybrid weather prediction model based on deep learning | |
CN116306951A (en) | Quantum computing method and device, electronic equipment and storage medium | |
Petrovic et al. | Firefly-xgboost approach for pedestrian detection |
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 | ||
GR01 | Patent grant |