CN-121981158-A - Target information missing value completion method, system, electronic equipment and storage medium based on generation of countermeasure network

CN121981158ACN 121981158 ACN121981158 ACN 121981158ACN-121981158-A

Abstract

The invention provides a target information missing value completion method, a target information missing value completion system, electronic equipment and a storage medium based on a generated countermeasure network. The method comprises the steps of S1, constructing a complement circulation neural network for processing missing time sequence target information, learning essential characteristics and missing rules of the missing time sequence target information through the complement circulation neural network, S2, searching a low-dimensional characteristic vector for each piece of missing time sequence target information in the complement circulation neural network by adopting a gradient descent algorithm, and S3, generating an optimal complement value in an countermeasure network based on the low-dimensional characteristic vector, and using the missing value in the complement target information. The method fully utilizes the dimension reduction capability of the noise reduction self-encoder, and automatically searches the corresponding low-dimension feature vector for each piece of missing target information. The missing value in the end-to-end automatic complement time sequence data can be achieved by combining the generation of the countermeasure network technology, and the time efficiency is higher.

Inventors

WU JIANYU
ZENG JIANGFENG
LI ZHE
LIU RULEI
CHEN ZHUO
HAN WEI
LIANG XU
LUO FUYU
LUO YE
DONG DING
LI PENGLIN
Xu Fengchi

Assignees

中国船舶集团有限公司系统工程研究院
中船智海创新研究院有限公司

Dates

Publication Date: 20260505
Application Date: 20251225
Priority Date: 20241230

Claims (10)

1. A target information missing value completion method based on generation of an antagonism network, the method comprising: s1, constructing a complement circulation neural network for processing the missing time sequence target information, and learning essential characteristics and missing rules of the missing time sequence target information through the complement circulation neural network; S2, searching a low-dimensional feature vector by adopting a gradient descent algorithm aiming at each piece of missing time sequence target information in the complement circulating neural network; And S3, generating an optimal complement value in the generation countermeasure network based on the low-dimensional feature vector, and complementing the missing value in the target information by using the complement value.
2. The method for supplementing target information missing values based on generation countermeasure network of claim 1, wherein in the step S1, the method for learning the essential features and missing rules of the missing time sequence target information through the complementary recurrent neural network comprises the step of coding the missing time sequence data through the historical memory vectors of the decaying time sequence target information by the complementary recurrent neural network so as to obtain the essential features and missing rules of the missing time sequence data.
3. The method according to claim 2, wherein in the step S1, for the data missing for a long time in a certain dimension, the history memory vector of the dimension is attenuated correspondingly according to the missing time, and the longer the missing time, the more the history memory vector is attenuated, and vice versa.
4. A method of supplementing a missing target information value based on generation of an countermeasure network according to claim 2 or 3, wherein in the step S1, in order to effectively attenuate the history memory vector of the recurrent neural network, an attenuation factor β is introduced, the calculation formula of which is as follows: wherein w β and b β are neural network parameters to be trained, e represents a natural constant, and delta ti represents the time interval of each dimension at the current moment; The attenuation factor beta is a core component of the complement loop nerve unit, and after the value of the attenuation factor beta is obtained, the complement loop nerve unit correspondingly attenuates the history memory vector according to the attenuation factor beta before updating the history memory vector in each iteration.
5. The method of claim 4, wherein in the step S1, an update formula of the complement recurrent neural unit based on the gated recurrent neural network is as follows: Where r ti is the reset gate, z ti is the update gate, For candidate history memory vectors, σ is a sigmoid function, Is the history memory vector of the last moment, Is the history memory vector of the last time after attenuation, Is a history-memory vector of the data, Is the x vector value at the current time, W xz is the W parameter after the superposition of the xz vector, b z is the b parameter under the action of the z vector, W xr is the W parameter after the superposition of the xr vector, W hr is the W parameter after the superposition of the hr vector, b r is the b parameter under the action of the r vector, W xh is the W parameter after the superposition of the xh vector, W hh is the W parameter under the influence of the h vector, b h is the b parameter under the action of the h vector, and W and b are parameters of the neural network.
6. A method of supplementing missing target information values based on a generation countermeasure network according to any one of claims 1 to 3, wherein in the step S3, the generation countermeasure network includes a generator G and a discriminator D, the input of the generator G is a low-dimensional random vector z, the output of the generator G is complete pseudo-time series data, the input of the discriminator D is divided into two parts, the first part is real missing time series data x, the second part is pseudo-but complete time series data G (z), and the output of the discriminator is a probability value that the two parts of samples are true.
7. The method for supplementing target information missing values based on a generation countermeasure network according to claim 6, wherein in the step S3, the generator G mainly comprises a supplementing circulating neural unit and a fully-connected layer neural network, an autoregressive strategy is adopted by the generator G when time series data are generated, firstly, an input vector Z is imported into the fully-connected layer neural network, the dimension of the output vector of the layer network is controlled through controlling the number of neurons of the fully-connected layer, so that the dimension of the output vector is consistent with the dimension of original time series data, then the output vector is imported into the supplementing circulating neural unit, the output vector of the supplementing circulating neural unit at the current moment is imported into the input of the same supplementing circulating neural unit at the next moment after the adjustment dimension of the fully-connected neural network, so that the self-iteration is continuously performed, and at the last moment, all the output vectors after the adjustment dimension of the fully-connected neural network are combined into a time series according to a sequential order, namely, the brand-new time series data generated by the generator G.
8. A target information missing value completion system based on generation of an countermeasure network, the system comprising: the first processing module is configured to construct a complement circulation neural network for processing the missing time sequence target information, and learn essential characteristics and missing rules of the missing time sequence target information through the complement circulation neural network; The second processing module is configured to search a low-dimensional feature vector by adopting a gradient descent algorithm aiming at each piece of missing time sequence target information in the complement circulating neural network; And the third processing module is configured to search a low-dimensional feature vector by adopting a gradient descent algorithm aiming at each piece of missing time sequence target information in the complement circulating neural network.
9. An electronic device comprising a memory storing a computer program and a processor implementing the steps of a target missing information value supplementing method according to any of claims 1 to 7 based on generating an antagonizing network when the computer program is executed.
10. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a computer program which, when executed by a processor, implements the steps in a target information deficiency value complementing method based on generation of an countermeasure network as claimed in any one of claims 1 to 7.

Description

Target information missing value completion method, system, electronic equipment and storage medium based on generation of countermeasure network The present application claims priority from China patent office, application No. 202411963911.4, application name "target information missing value completion method, system, electronic device and storage Medium based on generating countermeasure network", filed on 30 th 12 th 2024, the entire contents of which are incorporated herein by reference. Technical Field The invention belongs to the technical field of unmanned ship technology and navigation control, and particularly relates to a target information missing value complement method, system, electronic equipment and storage medium based on a generated countermeasure network. Background Unmanned ship (unmanned surface vessel, USV), which is an unmanned surface vessel, fills up science and technology and future mystery colors. As an intelligent product, the unmanned ship has excellent autonomy, can independently analyze and plan an optimal sailing path without depending on human beings, and greatly improves sailing efficiency. Meanwhile, the unmanned ship has obvious low risk, and safety accidents caused by negligence or misoperation of a driver can be effectively avoided. In addition, unmanned ship also has powerful environmental adaptation ability, and can accomplish the task with margin in both calm sea surface and rough sea. Although unmanned ship has very high application value when carrying out target measurement task, because of the restriction of self range and the restriction of communication distance, hardly receive artificial control constantly in the operation process. Due to possible instability or interference of various sensors of the unmanned ship, the acquired time sequence data are often incomplete, and the missing part of the data prevents deep analysis of target information. When the missing value is complemented, time sequence information in the target information is further considered, so that an accurate complement effect is obtained. The generation of a countermeasure Network (GAN) in 2014 provides a possibility for the problem of data loss in complex scenarios. Through development in recent years, a superior derivative model is iterated to generate a antagonism network, and students gradually apply the model to the unmanned ship data filling field. Generating the challenge network is a generating model (GENERATIVE MODEL) whose underlying basic idea is to obtain a number of training samples (Training Examples) from a training library, thereby learning the probability distribution generated by the training cases. Some generative models may give an estimate of the probability distribution function definition, while others may give new samples from the probability distribution of the original generated training library. However, when the conventional recurrent neural network processes the missing time sequence target information, the longer the data missing time is, the more the history memory vector should be attenuated, so that a method for accelerating the completion of the missing value in the target information must be designed to complete the missing value of the target information. The technical problems are to be solved. Disclosure of Invention In order to solve the technical problems, the invention provides a technical scheme based on a target information missing value completion method for generating an antagonism network, so as to solve the technical problems. The first aspect of the invention discloses a target information missing value completion method based on generation of an antagonism network, which comprises the following steps: S1, constructing a complement circulation neural network for processing the missing time sequence target information, and learning essential characteristics and missing rules of the missing time sequence target information through the complement circulation neural network; S2, searching a low-dimensional feature vector by adopting a gradient descent algorithm aiming at each piece of missing time sequence target information in the complement circulating neural network; and S3, generating an optimal complement value in the generation countermeasure network based on the low-dimensional feature vector, and complementing the missing value in the target information by using the complement value. According to the method of the first aspect of the invention, in step S1, the method for learning the essential characteristics and the missing rule of the missing time sequence target information through the complement circulating neural network comprises the steps of encoding the missing time sequence data through the historical memory vector of the decaying time sequence target information by the complement circulating neural network, so as to obtain the essential characteristics and the missing rule of the missing time sequence data. According to the metho