UUV power system fault prediction method, device and equipment and storage medium

Abstract

The invention provides a UUV power system fault prediction method, device and equipment and a storage medium, and relates to the technical field of unmanned underwater vehicles, and the method comprises the steps: obtaining a feature matrix of a UUV power system at a plurality of continuous moments and an adjacent matrix corresponding to the feature matrix, the adjacent matrix being used for representing an influence relation between parameters of the UUV power system; sequentially inputting the adjacent matrix and the feature matrixes at the plurality of continuous moments into a feature extraction layer to obtain a plurality of first matrixes; and determining a fault prediction result based on the plurality of first matrixes. Through the arrangement, the fault prediction result is determined based on the multiple features of the power system at the multiple continuous moments and the influence relation among the features, the fault prediction accuracy is improved, potential faults can be found more accurately, damage of sudden faults to equipment is reduced, the operation reliability degree in the task execution process is improved, and the task execution efficiency is improved. And maintenance according to conditions can be more accurately realized, and the maintenance efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical field of underwater unmanned vehicles, and particularly to a method, device, equipment and storage medium for predicting faults in a UUV power system. Background Art

[0002] Fault prediction technology has extremely important significance and far-reaching influence in the technical field of underwater unmanned vehicles (UUVs). When a UUV is performing a mission, the reliability of the power system is directly related to the success or failure of the mission. Through fault prediction technology, potential faults in the power system can be detected, damage to the equipment caused by sudden faults can be reduced, thereby extending the service life of the UUV. On the other hand, it can participate in intelligent decision-making, thereby improving the mission effectiveness and safety of the UUV. Through fault prediction, condition-based maintenance can be achieved, unnecessary maintenance activities can be reduced, maintenance costs can be lowered, and maintenance efficiency can be improved.

[0003] However, currently, the prediction methods for faults in the power system of UUVs have poor accuracy. Summary of the Invention

[0004] The present application provides a method, device, equipment and storage medium for predicting faults in a UUV power system, which can determine a fault prediction result based on multiple characteristics of the UUV power system at multiple consecutive moments and the influence relationship between the characteristics, improving the accuracy of fault prediction.

[0005] To achieve the above object, the present application adopts the following technical solutions: In a first aspect, the present application provides a method for predicting faults in a UUV power system, the method comprising: Obtaining a feature matrix of the UUV power system at multiple consecutive moments and an adjacency matrix corresponding to the feature matrix, the adjacency matrix being used to represent the influence relationship between the parameters of the UUV power system; Successively inputting the adjacency matrix and the feature matrices at multiple consecutive moments into a feature extraction layer to obtain multiple first matrices; Determining a fault prediction result based on the multiple first matrices.

[0006] In some possible implementation manners, the feature extraction layer includes a first feature extraction layer and a second feature extraction layer, and the feature extraction methods of the first feature extraction layer and the second feature extraction layer are different; the successively inputting the adjacency matrix and the feature matrices at multiple consecutive moments into the feature extraction layer to obtain multiple first matrices includes: Inputting the adjacency matrix and the feature matrices at multiple consecutive moments into the first feature extraction layer to obtain multiple first sub-matrices; Input the adjacency matrix and the feature matrices at multiple consecutive time instants into the second feature extraction layer to obtain multiple second sub-matrices; Fuse the multiple first sub-matrices and the multiple second sub-matrices according to each time instant to obtain multiple first matrices.

[0007] In some possible implementation manners, determining the fault prediction result based on the multiple first matrices includes: Input the multiple first matrices into a long short-term memory network (LSTM) to obtain a second matrix; Obtain the fault prediction result according to the second matrix and a classifier.

[0008] In some possible implementation manners, obtaining the fault prediction result according to the second matrix and the classifier includes: Perform feature screening on the second matrix to obtain a third matrix; Input the third matrix into the classifier to obtain the fault prediction result.

[0009] In some possible implementation manners, inputting the adjacency matrix and the feature matrices at multiple consecutive time instants into the second feature extraction layer to obtain multiple second sub-matrices includes: Determine query vectors, key vectors, and value vectors corresponding to respective eigenvalues at multiple consecutive time instants according to the feature matrices at multiple consecutive time instants; Determine the association relationships between the respective eigenvalues according to the adjacency matrix; Determine weighted eigenvalues corresponding to the respective eigenvalues at multiple consecutive time instants according to the query vectors, key vectors, and value vectors of the respective eigenvalues at multiple consecutive time instants and the association relationships between the respective eigenvalues; Concatenate the weighted eigenvalues corresponding to the respective eigenvalues at multiple consecutive time instants in sequence to obtain multiple second sub-matrices.

[0010] In some possible implementation manners, the features of the UUV power system include: motor speed, motor power, motor temperature, controller temperature, torque, rotor position offset, power input of the power battery, cabin temperature, cabin humidity, and cabin atmospheric pressure.

[0011] In a second aspect, the present application provides a UUV power system fault prediction device, and the device includes: A feature construction module, configured to obtain a feature matrix of the UUV power system at multiple consecutive time instants and an adjacency matrix corresponding to the feature matrix, where the adjacency matrix is used to represent the influence relationship between parameters of the UUV power system; A feature extraction module, configured to sequentially input the adjacency matrix and the feature matrices at multiple consecutive time instants into a feature extraction layer to obtain multiple first matrices; A prediction module, configured to determine a fault prediction result based on a plurality of first matrices.

[0012] In some possible implementation manners, the feature extraction layer includes a first feature extraction layer and a second feature extraction layer, and the feature extraction manners of the first feature extraction layer and the second feature extraction layer are different; specifically, the feature extraction module is configured to input an adjacency matrix and feature matrices at multiple consecutive moments into the first feature extraction layer to obtain a plurality of first sub-matrices; input the adjacency matrix and feature matrices at multiple consecutive moments into the second feature extraction layer to obtain a plurality of second sub-matrices; and fuse the plurality of first sub-matrices and the plurality of second sub-matrices according to each moment to obtain a plurality of first matrices.

[0013] In a third aspect, the present application provides a computing device, which includes a memory and a processor; wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the computing device is enabled to execute the method as described in the first aspect.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium, which is used to store a computer program, and the computer program is used to execute the method as described in the first aspect.

[0015] It can be seen from the above technical solutions that the present application has at least the following beneficial effects: The present application provides a method for fault prediction of a UUV power system. This method can be executed by a processing device. The processing device obtains a feature matrix composed of a plurality of features such as motor temperature, power battery input power, and environmental temperature of the power system at multiple consecutive moments, and uses an adjacency matrix to construct the association between the plurality of features, that is, obtains the influence relationship between the plurality of features. Based on the plurality of features of the UUV power system at multiple consecutive moments and the influence relationship between the features, a fault prediction result is determined, which improves the accuracy of fault prediction, is beneficial to more accurately discovering potential faults, reducing the damage of sudden faults to equipment, improving the operation reliability of the UUV during the mission execution process, and is also beneficial to more accurately realizing condition-based maintenance, reducing unnecessary maintenance activities, reducing maintenance costs, and improving maintenance efficiency.

[0016] It should be understood that the description of technical features, technical solutions, beneficial effects or similar language in this application does not imply that all features and advantages can be realized in any single embodiment. On the contrary, it is understood that the description of features or beneficial effects means that specific technical features, technical solutions or beneficial effects are included in at least one embodiment. Therefore, the description of technical features, technical solutions or beneficial effects in this specification does not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions and beneficial effects described in the present embodiment can also be combined in any appropriate manner. Those skilled in the art will understand that the embodiment can be realized without one or more specific technical features, technical solutions or beneficial effects of a specific embodiment. In other embodiments, additional technical features and beneficial effects can also be identified in a specific embodiment that does not embody all embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of a UUV power system fault prediction method provided in an embodiment of the present application; Figure 2 A schematic diagram of the influencing relationship between multiple features of the UUV power system provided in an embodiment of the present application; Figure 3 A schematic diagram of an adjacency matrix formed by multiple features of a UUV power system provided in an embodiment of the present application; Figure 4 A schematic diagram of the structure of a long short-term memory network model provided in an embodiment of the present application; Figure 5 A schematic diagram of a UUV power system fault prediction device provided in an embodiment of the present application; Figure 6 A schematic diagram of a computing device provided for an embodiment of the present application. DETAILED DESCRIPTION

[0018] The terms "first", "second", "third", etc. in the specification of this application and the accompanying drawings are used to distinguish different objects rather than to limit a specific order.

[0019] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0020] In order to make the description of the following embodiments clear and concise, a brief introduction to the related technology is first given: Fault prediction technology is of extremely important significance and far-reaching impact in both military and civilian Unmanned Underwater Vehicles (UUVs). In the military field, when a UUV is performing a mission, the reliability of its propulsion system is directly related to the success or failure of the mission. Fault prediction technology can detect potential faults in advance, reduce the damage to equipment caused by sudden faults, thereby extending the service life of the UUV. On the other hand, it can participate in intelligent decision-making, thus improving the mission effectiveness and safety of the UUV. Through fault prediction, condition-based maintenance can be achieved, unnecessary maintenance activities can be reduced, maintenance costs can be lowered, and maintenance efficiency can be improved, which is of great significance for improving the modern management level of equipment.

[0021] In the civilian field, fault prediction of UUVs can reduce production interruptions and maintenance costs caused by faults, and improve economic benefits. When civilian UUVs are performing tasks such as submarine pipeline inspection and environmental monitoring, their safety is directly related to public safety. Fault prediction technology can prevent the occurrence of major accidents and ensure public safety.

[0022] Existing UUV fault prediction is only based on a single factor for fault prediction. For example, only based on the output power of the battery to predict battery body faults or faults in the electric power system, and only based on the motor speed to predict motor bearing faults or motor winding faults, but the mutual influence between various factors is not considered, resulting in inaccurate fault prediction.

[0023] In view of this, the embodiment of this application provides a UUV power system fault prediction method. This method can be executed by a processing device. The processing device obtains a feature matrix composed of multiple features such as motor temperature, power battery input power, and environmental temperature at multiple consecutive moments of the power system, constructs the association between multiple features using an adjacency matrix, that is, obtains the influence relationship between multiple features, and determines the fault prediction result based on multiple features of the UUV power system at multiple consecutive moments and the influence relationship between features, improving the accuracy of fault prediction, facilitating more accurate discovery of potential faults, reducing the damage to equipment caused by sudden faults, improving the operating reliability of the UUV during mission execution, and also facilitating more accurate implementation of condition-based maintenance, reducing unnecessary maintenance activities, lowering maintenance costs, and improving maintenance efficiency.

[0024] The application scenario of the UUV power system fault prediction method is introduced below: For example, in the military field, it is applied during the mission of UUVs for military reconnaissance and anti-submarine warfare; or in the civilian field, it is applied during the mission of UUVs for submarine pipeline inspection, environmental monitoring, etc.

[0025] The following combines Figure 1 The flowchart of the prediction method shown to introduce this prediction method. This prediction method includes: S101. The processing device obtains the feature matrix of the UUV power system at multiple consecutive moments, as well as the corresponding adjacency matrix, where the adjacency matrix is used to represent the influence relationship between the parameters of the UUV power system.

[0026] The UUV power system is an integrated system that provides propulsion power, energy supply, and control for the UUV. The UUV power system can ensure underwater autonomous navigation, mission execution, and long-term operation. The UUV power system includes an energy supply module, a power module, a thermal management module, etc.

[0027] The features of the UUV power system include motor speed, motor power, motor temperature, motor controller temperature, torque, rotor position offset, power battery input power, cabin temperature, cabin humidity, and cabin atmospheric pressure.

[0028] Sensors are set in the UUV power system, and each of the above features is collected by a corresponding sensor. For example, the motor temperature can be collected by a correspondingly set temperature sensor, and the cabin atmospheric pressure can be collected by a correspondingly set pressure sensor. The sensor transmits the detection signal to the processing device, and the processing device converts the detection signal into an initial feature parameter and stores it.

[0029] Among them, the sensor collects data at a set time interval, for example, collects data once every 1 minute.

[0030] The processing device processes each initial feature parameter collected by the sensor at each moment to obtain each target feature parameter at each moment.

[0031] Specifically, data cleaning is performed on each initial feature parameter to obtain multiple remaining feature parameters; normalization is performed on each remaining feature parameter to obtain each target feature parameter.

[0032] Data cleaning is specifically to filter out obvious noise data. The noise data is, for example: noise data such as the power battery input power being less than 0, the motor speed being less than 0, the motor temperature being lower than the cabin temperature or exceeding the maximum temperature, the motor controller temperature being lower than the cabin temperature or exceeding the maximum temperature, the torque being less than 0, and the mutation of each index, etc., to minimize the impact of noise data on the accuracy of the network model constructed based on the feature parameters.

[0033] Each remaining feature parameter can be normalized through the following formula to obtain each target feature parameter:

[0034] Where represents the remaining feature parameter, , respectively represent the maximum value and the minimum value of the feature corresponding to the remaining feature parameter, Represents the target characteristic parameter.

[0035] For example, perform data cleaning on the motor speed in the initial characteristic parameters, filter out the noise data where the motor speed is less than 0; normalize the remaining motor speeds according to the above formula to obtain the target characteristic parameter.

[0036] In this embodiment, the normalization of the data converts data with different dimensions to the same dimension, which is beneficial to improving the accuracy of the network model constructed based on the characteristic parameters and the convergence speed during training.

[0037] The processing device annotates each target characteristic parameter at each moment according to the historical storage data and relevant information of the maintenance and protection records in chronological order. If the power system fails at the corresponding moment, each target characteristic parameter at that moment is annotated as 1; if the power system does not fail at the corresponding moment, each target characteristic parameter at that moment is annotated as 0. After the annotation is completed, the subsequent Figure 4 shown network model can be trained to achieve fault prediction.

[0038] The processing device arranges each target characteristic parameter at each moment to obtain a 1×n characteristic matrix corresponding to each moment, and then splices the characteristic matrices of each moment in chronological order to obtain a three-dimensional characteristic matrix of 1×n×T, where n represents the number of characteristics of the power system and T represents the length of the time series.

[0039] Construct an adjacency matrix according to the mutual influence between the characteristics of the UUV power system.

[0040] Represent each characteristic of the UUV power system with a corresponding symbol, and the corresponding relationship between the characteristics and the symbols is shown in Table 1 below.

[0041] Table 1:

[0042] Analyze the mutual influence between the characteristics in Table 1 through graph relationship. As Figure 2 shown, regard the mutual influence between the characteristics as a directed graph, where the nodes represent the characteristics and the edges represent the influence. For example, the motor power and the motor temperature have mutual influence, then Figure 2 there is a two-way arrow between and ; the motor temperature and the motor controller temperature Figure 2 do not have mutual influence, then there is no There is no bidirectional arrow between them.

[0043] If the feature For the feature There is an interaction between them, then the adjacency matrix If the feature For the feature There is no interaction between them, then . Among them, Represents the th feature, Represents the th feature. As Figure 3 Shown is the adjacency matrix composed of the feature set in Table 1. For example, there is an interaction between the motor power and the motor temperature , then the corresponding Figure 3 in ; there is no interaction between the motor temperature and the motor controller temperature , then the corresponding Figure 3 in .

[0044] S102. The processing device sequentially inputs the adjacency matrix and the feature matrices at multiple consecutive time instants into the feature extraction layer to obtain multiple first matrices.

[0045] In a specific embodiment, the feature extraction layer includes a first feature extraction layer and a second feature extraction layer, and the feature extraction methods of the first feature extraction layer and the second feature extraction layer are different.

[0046] The method of sequentially inputting the adjacency matrix and the feature matrices at multiple consecutive time instants into the feature extraction layer to obtain multiple first matrices includes: Input the adjacency matrix and the feature matrices at multiple consecutive time instants into the first feature extraction layer to obtain multiple first sub-matrices.

[0047] Specifically, referring to Figure 4 the schematic diagram of the network model architecture shown, the first feature extraction layer is a graph convolutional layer (GCN, Graph Convolutional Network). The graph convolutional layer is a neural network layer specifically used to process graph-structured data and can effectively capture the topological relationship between nodes.

[0048] The first sub-matrix can be obtained through the following formula:

[0049] Among them, Represents the t-th first sub-matrix, Represents the adjacency matrix, Denote the adjacency matrix as the degree matrix, denote the feature matrix at the t-th moment, denote the parameter matrix of the first feature extraction layer; denote the activation function.

[0050] The degree matrix is a diagonal matrix, and its diagonal elements represent the degree of each node in the graph structure data, which is used to normalize the adjacency matrix.

[0051] The parameter matrix of the first feature extraction layer is a trainable weight matrix in the neural network, which is used to perform a linear transformation (feature extraction) on the feature matrix at the t-th moment.

[0052] Input the adjacency matrix and the feature matrices at multiple consecutive moments into the second feature extraction layer to obtain multiple second sub-matrices.

[0053] In a specific embodiment, referring to Figure 4 the schematic diagram of the network model architecture shown, the second feature extraction layer is a Transformer (self-attention model) layer. The Transformer layer is a deep learning model structure based on the self-attention mechanism.

[0054] The ways of inputting the adjacency matrix and the feature matrices at multiple consecutive moments into the second feature extraction layer to obtain multiple second sub-matrices include: Determine the query vectors, key vectors, and value vectors corresponding to the respective eigenvalues at multiple consecutive moments according to the feature matrices at multiple consecutive moments.

[0055] Specifically, the respective eigenvalues at multiple consecutive moments respectively undergo graph self-attention operations to respectively obtain a patch. Each patch represents a local topological structure, and each patch creates three vectors, namely Query (query vector), Key (key vector), and Value (value vector), in the way of a fully connected layer, denoted as the Q vector, K vector, and V vector.

[0056] Determine the correlation relationships between the respective eigenvalues according to the adjacency matrix.

[0057] If the feature has an interaction with the feature , then the adjacency matrix , if the feature has no interaction with the feature , then .

[0058] Determine the weighted eigenvalues corresponding to each eigenvalue at multiple consecutive moments according to the query vectors, key vectors, and value vectors of each eigenvalue at multiple consecutive moments and the association relationships between the eigenvalues.

[0059] Specifically, taking the feature as the feature , and the feature as other features except as an example, taking the calculation of the weighted eigenvalue corresponding to the motor speed as an example to illustrate: After the graph self-attention operation, is obtained, where the query vector, key vector, and value vector corresponding to are respectively represented as , , ; After the graph self-attention operation, is obtained, where the query vector, key vector, and value vector corresponding to are respectively represented as , , .

[0060] Express the weight relationship between and as .

[0061] When the motor speed has an impact on the feature , that is, , then calculate the weight between ; when the motor speed has no impact on the feature , that is, , then do not calculate.

[0062] At the t-th moment, the weight between and can be calculated using the following formula:

[0063] where represents the query vector corresponding to , represents the key vector corresponding to , represents the vector length of the key vector , represents the value vector corresponding to , Represents a normalization function.

[0064] For example, the motor speed on the motor power has an impact, then and The weight between them is calculated using the following formula:

[0065] Among them, represents and The weight between them, represents The corresponding query vector, represents The corresponding key vector, represents the key vector The vector length of, represents The corresponding value vector, t represents the t-th moment.

[0066] When calculating the motor speed and the features that have an impact on the motor speed After the weight between them, and The weight values between them are added according to the following formula to obtain the feature The weighted eigenvalue of :

[0067] Similarly, it is calculated that The weighted eigenvalues corresponding to these features respectively, for example, the weighted eigenvalue corresponding to the feature is represented as and the feature The weighted eigenvalue corresponding to is represented as .

[0068] For the weighted eigenvalues corresponding to each eigenvalue at multiple consecutive moments, they are concatenated in order to obtain multiple second sub-matrices.

[0069] At the t-th moment, for The weighted eigenvalues corresponding to these features are concatenated in order to obtain the corresponding second sub-matrix, which can be calculated through the following formula:

[0070] represents the t-th second sub-matrix, represents the number of UUV power system features, represents concatenation.

[0071] At each moment, a plurality of first sub-matrices and a plurality of second sub-matrices are fused to obtain a plurality of first matrices.

[0072] Specifically, a plurality of first sub-matrices and a plurality of second sub-matrices can be added and fused to obtain a plurality of first matrices.

[0073] In this embodiment, the graph convolutional layer extracts local structural features through the adjacency matrix. The graph convolutional layer captures the spatial topological dependencies of the features of the UUV power system based on the adjacency matrix, effectively modeling the fault propagation path; and guides the operation of self-attention in the Transformer layer through the adjacency matrix, thereby simulating the interaction relationship between features. This embodiment simultaneously uses the graph convolutional and Transformer structures guided by the adjacency matrix to ensure the correlation features between the extracted parameters, and can more accurately capture the coupling effect and collaborative anomalies among the feature parameters of the UUV power system, significantly improving the accuracy of fault prediction and the early warning ability.

[0074] S103. The processing device determines a fault prediction result based on the plurality of first matrices.

[0075] Specifically, referring to Figure 4 the schematic diagram of the network model architecture shown, the plurality of first matrices are input into the long short-term memory network (LSTM) to obtain a second matrix.

[0076] The long short-term memory network (LSTM) is a type of recurrent neural network with powerful time series analysis capabilities. It captures the trend of the UUV power system device state changing over time and can learn the relationship between these trends and faults.

[0077] The long short-term memory network (LSTM) extracts time relationship features. After being processed by the long short-term memory network (LSTM), features that integrate both time and space dimensions are obtained. Then, average pooling is used to obtain a feature vector. The calculation formula for the second matrix is as follows:

[0078] Among them, represents the calculation process of the long short-term memory network (LSTM), represents average pooling, represents the first matrix, represents the second matrix.

[0079] Based on the second matrix and the classifier, a fault prediction result is obtained.

[0080] Specifically, feature screening is performed on the second matrix to obtain a third matrix.

[0081] Perform feature screening on the second matrix to control the output of valid information and discard invalid information. Specifically, it can be achieved by using an activation function to perform data mapping. Map the second matrix through the activation function into a real number vector from 0 to 1, and multiply the result after mapping with the original feature vector bit by bit to weight each eigenvalue, achieving the purpose of screening valid eigenvalues. The specific calculation formula is as follows:

[0082] denotes the second matrix, denotes the third matrix.

[0083] Input the third matrix into the classifier to obtain the fault prediction result.

[0084] Specifically, input the third matrix into the classifier. Specifically, it can use as the classifier. The calculation formula for obtaining the fault prediction result is as follows:

[0085] denotes the fault prediction result, denotes the classification function, denotes the third matrix.

[0086] Using as the classifier, the obtained fault prediction result has an intuitive probability and good optimization effect when combined with the cross-entropy loss.

[0087] In this embodiment, input the first matrix at multiple moments into the LSTM to extract the cross-moment fault evolution pattern, making up for the locality limitation of GCN and Transformer in long-sequence modeling and improving the time resolution of fault detection; perform Sigmoid weighted screening on the second matrix output by the LSTM before classification to retain high-contribution features; input the screened third matrix into the classifier to reduce the risk of overfitting and improve the generalization ability of the model in the data imbalance scenario.

[0088] The UUV power system fault prediction method provided by the embodiments of the present application can be executed by a processing device. The processing device obtains a feature matrix composed of multiple features such as motor temperature, power battery input power, and ambient temperature at multiple consecutive moments of the power system, constructs the association between multiple features by using an adjacency matrix, that is, obtains the influence relationship between multiple features, and determines the fault prediction result based on multiple features and the influence relationship between features at multiple consecutive moments of the UUV power system, improving the accuracy of fault prediction, facilitating more accurate discovery of potential faults, reducing the damage of sudden faults to equipment, improving the operating reliability of the UUV during mission execution, and also facilitating more accurate implementation of condition-based maintenance, reducing unnecessary maintenance activities, reducing maintenance costs, and improving maintenance efficiency.

[0089] In a specific embodiment, the influence relationship between features is constructed based on a graph convolutional layer, the temporal relationship between features is constructed by a Transformer layer, and an LSTM network is used to extract spatio-temporal features, thereby obtaining a complete network model.

[0090] The processed standard formatted spatio-temporal data is input into the network model for training to obtain an optimal model. The fault information to be detected is input into the trained optimal model for fault prediction.

[0091] As described above in conjunction with Figures 1 to 4 the UUV power system fault prediction method provided by the embodiments of the present application has been introduced in detail. Next, the devices and equipment provided by the embodiments of the present application will be introduced with reference to the accompanying drawings.

[0092] As Figure 5 shown, this figure is a schematic diagram of a UUV power system fault prediction device provided by the embodiments of the present application. The device includes: A feature construction module 301, configured to obtain a feature matrix of the UUV power system at multiple consecutive moments and an adjacency matrix corresponding to the UUV power system, where the adjacency matrix is used to represent the influence relationship between the parameters of the UUV power system.

[0093] A feature extraction module 302, configured to sequentially input the adjacency matrix and the feature matrix at multiple consecutive moments into a feature extraction layer to obtain multiple first matrices.

[0094] A prediction module 303, configured to determine a fault prediction result based on multiple first matrices.

[0095] Optionally, the feature extraction layer includes a first feature extraction layer and a second feature extraction layer, and the feature extraction methods of the first feature extraction layer and the second feature extraction layer are different; the feature extraction module 302 is specifically configured to input the adjacency matrix and the feature matrices at multiple consecutive time instants into the first feature extraction layer to obtain multiple first sub-matrices; input the adjacency matrix and the feature matrices at multiple consecutive time instants into the second feature extraction layer to obtain multiple second sub-matrices; and fuse the multiple first sub-matrices and the multiple second sub-matrices according to each time instant to obtain multiple first matrices.

[0096] Optionally, the feature extraction module 302 is specifically configured to determine query vectors, key vectors, and value vectors corresponding to respective eigenvalues at multiple consecutive time instants according to the feature matrices at multiple consecutive time instants; determine the association relationships between the respective eigenvalues according to the adjacency matrix; determine weighted eigenvalues corresponding to the respective eigenvalues at multiple consecutive time instants according to the query vectors, key vectors, and value vectors of the respective eigenvalues at multiple consecutive time instants and the association relationships between the respective eigenvalues; and concatenate the weighted eigenvalues corresponding to the respective eigenvalues at multiple consecutive time instants in sequence to obtain multiple second sub-matrices.

[0097] Optionally, the prediction module 303 is specifically configured to input the multiple first matrices into a long short-term memory network (LSTM) to obtain a second matrix; and obtain a fault prediction result according to the second matrix and a classifier.

[0098] Optionally, the prediction module 303 is specifically configured to perform feature screening on the second matrix to obtain a third matrix; and input the third matrix into a classifier to obtain a fault prediction result.

[0099] The embodiment of the present application further provides a computing device, as Figure 6 shown in the figure which is a schematic diagram of a computing device provided by the embodiment of the present application. The computing device 200 includes a bus 201, a processor 202, a communication interface 203, and a memory 204. The processor 202, the memory 204, and the communication interface 203 communicate with each other through the bus 201.

[0100] The bus 201 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For the sake of simplicity of representation, Figure 6 only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.

[0101] The processor 202 can be any one or more of processors such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0102] The communication interface 203 is used for external communication.

[0103] The memory 204 may include a volatile memory, such as a random access memory (RAM). The memory 204 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0104] Executable code is stored in the memory 204, and the processor 202 executes the executable code to perform the foregoing virtual object allocation method.

[0105] Specifically, in the case of implementing Figure 5 the illustrated embodiment, and Figure 5 when each module or unit of the UUV power system fault prediction device described in the embodiment is implemented by software, the software or program code required to execute the functions of each module / unit in Figure 5 can be partially or fully stored in the memory 204. The processor 202 executes the program code corresponding to each unit stored in the memory 204 to perform the foregoing UUV power system fault prediction method.

[0106] An embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computing device can store or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, or a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state drive), etc. The computer-readable storage medium includes instructions that direct the computing device to execute the foregoing virtual object allocation method.

[0107] The embodiments of the present application also provide a computer program product, which includes one or more computer instructions. When the computer instructions are loaded and executed on a computing device, they wholly or partly generate the processes or functions according to the embodiments of the present application.

[0108] The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from a website, a computer, or a data center to another website, a computer, or a data center in a wired manner (such as coaxial cable, optical fiber, digital subscriber line) or a wireless manner (such as infrared, wireless, microwave, etc.).

[0109] When the computer program product is executed by a computer, the computer executes any one of the aforementioned methods for predicting UUV power system faults. The computer program product may be a software installation package. In the case where any one of the aforementioned methods for predicting UUV power system faults is needed, the computer program product may be downloaded and executed on the computer.

[0110] The descriptions of the processes or structures corresponding to the above respective drawings each have their own focuses. For parts not detailed in a certain process or structure, reference may be made to the relevant descriptions of other processes or structures.

[0111] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present application shall be covered by the protection scope of the present application.

Claims

1. A method for predicting faults in a UUV power system, characterized in that, The method includes: Obtaining a feature matrix of the UUV power system at multiple consecutive moments and an adjacency matrix corresponding to the feature matrix, where the adjacency matrix is used to represent the influence relationship between the parameters of the UUV power system; Sequentially inputting the adjacency matrix and the feature matrices at multiple consecutive moments into a feature extraction layer to obtain multiple first matrices; Determining a fault prediction result based on the multiple first matrices.

2. The method according to claim 1, characterized in that, The feature extraction layer includes a first feature extraction layer and a second feature extraction layer, and the feature extraction methods of the first feature extraction layer and the second feature extraction layer are different; The sequentially inputting the adjacency matrix and the feature matrices at multiple consecutive moments into the feature extraction layer to obtain multiple first matrices includes: Inputting the adjacency matrix and the feature matrices at multiple consecutive moments into the first feature extraction layer to obtain multiple first sub-matrices; Inputting the adjacency matrix and the feature matrices at multiple consecutive moments into the second feature extraction layer to obtain multiple second sub-matrices; Fusing the multiple first sub-matrices and the multiple second sub-matrices according to each moment to obtain multiple first matrices.

3. The method according to claim 1, characterized in that The determining a fault prediction result based on the multiple first matrices includes: Inputting the multiple first matrices into a long short-term memory network (LSTM) to obtain a second matrix; Obtaining a fault prediction result according to the second matrix and a classifier.

4. The method according to claim 3, wherein The obtaining a fault prediction result according to the second matrix and the classifier includes: Performing feature screening on the second matrix to obtain a third matrix; Inputting the third matrix into the classifier to obtain a fault prediction result.

5. The method according to claim 2, wherein Inputting the adjacency matrix and the feature matrices at multiple consecutive moments into the second feature extraction layer to obtain multiple second sub-matrices includes: Determining query vectors, key vectors, and value vectors corresponding to each eigenvalue at multiple consecutive moments according to the feature matrices at multiple consecutive moments; Determining the correlation relationship between each eigenvalue according to the adjacency matrix; Determining weighted eigenvalues corresponding to each eigenvalue at multiple consecutive moments according to the query vectors, key vectors, and value vectors of each eigenvalue at multiple consecutive moments and the correlation relationship between each eigenvalue; Sequentially splicing the weighted eigenvalues corresponding to each eigenvalue at multiple consecutive moments to obtain multiple second sub-matrices.

6. The method according to claim 1, characterized in that, The features of the UUV power system include: motor speed, motor power, motor temperature, motor controller temperature, torque, rotor position offset, power battery input power, cabin temperature, cabin humidity, and cabin atmospheric pressure.

7. A fault prediction device for a UUV power system, characterized in that, The device includes: A feature construction module for obtaining a feature matrix of the UUV power system at multiple consecutive moments and an adjacency matrix corresponding to the feature matrix, where the adjacency matrix is used to represent the influence relationship between the parameters of the UUV power system; A feature extraction module for sequentially inputting the adjacency matrix and the feature matrices at multiple consecutive moments into a feature extraction layer to obtain multiple first matrices; A prediction module for determining a fault prediction result based on the multiple first matrices.

8. The device according to claim 7, wherein The feature extraction layer includes a first feature extraction layer and a second feature extraction layer, and the feature extraction methods of the first feature extraction layer and the second feature extraction layer are different; the feature extraction module is specifically configured to input the adjacency matrix and the feature matrices at multiple consecutive time instants into the first feature extraction layer to obtain a plurality of first sub-matrices; input the adjacency matrix and the feature matrices at multiple consecutive time instants into the second feature extraction layer to obtain a plurality of second sub-matrices; and fuse the plurality of first sub-matrices and the plurality of second sub-matrices according to each time instant to obtain a plurality of first matrices.

9. A computing device, characterized in that, comprising a memory and a processor; Wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the computing device is caused to execute the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 6.

Images