WO2024073936A1 - 一种适用于极端状态的储能设备的故障预测模型生成方法 - Google Patents

一种适用于极端状态的储能设备的故障预测模型生成方法 Download PDF

Info

Publication number
WO2024073936A1
WO2024073936A1 PCT/CN2022/137740 CN2022137740W WO2024073936A1 WO 2024073936 A1 WO2024073936 A1 WO 2024073936A1 CN 2022137740 W CN2022137740 W CN 2022137740W WO 2024073936 A1 WO2024073936 A1 WO 2024073936A1
Authority
WO
WIPO (PCT)
Prior art keywords
fault
operation data
data
extreme
prediction model
Prior art date
Application number
PCT/CN2022/137740
Other languages
English (en)
French (fr)
Inventor
杨之乐
安钊
郭媛君
刘祥飞
江俊杰
Original Assignee
深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2024073936A1 publication Critical patent/WO2024073936A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding

Definitions

  • the present invention relates to the field of model generation, and in particular to a method for generating a fault prediction model for energy storage equipment in extreme states.
  • Traditional energy storage equipment fault monitoring mainly relies on professional staff to monitor the operating data of energy storage equipment in real time, which requires a lot of manpower. And because manual monitoring has a certain degree of subjectivity, when the staff is inexperienced or does not understand the current energy storage equipment well, it is difficult to accurately predict the failure of the energy storage equipment, thus affecting the normal operation of the energy storage system.
  • fault prediction models based on machine learning have gradually replaced manual monitoring.
  • the fault prediction model is to judge the equipment status reflected by future parameters by learning the characteristics of historical parameters. Therefore, the fault prediction model requires a large amount of historical data for training. However, it is difficult to obtain historical fault data of energy storage equipment in extreme environments, and the number of acquisitions is small.
  • the technical problem to be solved by the present invention is that, in view of the above-mentioned defects of the prior art, a method for generating a fault prediction model suitable for energy storage devices in extreme conditions is provided, aiming to solve the problem that the existing fault prediction models are difficult to apply to the fault prediction of energy storage devices in extreme environments due to the lack of sufficient historical failure data of energy storage devices in extreme environments.
  • an embodiment of the present invention provides a method for generating a fault prediction model for an energy storage device in an extreme state, wherein the method comprises:
  • the extreme fault operation data is used to reflect the operation data of the target device when a fault occurs in an extreme environment, and the extreme environment is an environment in which the acquisition frequency is lower than a preset threshold;
  • the fault prediction model is trained according to the first training data set, and a target fault prediction model corresponding to the target device is determined according to the trained fault prediction model.
  • each of the extreme fault operation data is time series data
  • each of the time series data includes a plurality of elements
  • the plurality of elements are respectively used to reflect the operation values corresponding to the same operation parameter at different time points.
  • the restoration of each of the extreme fault operation data to obtain the restored fault operation data corresponding to each of the extreme fault operation data includes:
  • the sequence data to be restored are restored to obtain the restored fault operation data corresponding to each time series data.
  • the restoring each of the to-be-restored sequence data to obtain the restored fault operation data corresponding to each of the time series data includes:
  • the restored fault operation data corresponding to each of the fault operation data to be restored outputted by the target prediction network based on each of the fault operation data to be restored are obtained.
  • the training process of the target prediction network includes:
  • the second training data set includes a number of missing operating data and a number of complete operating data corresponding to the missing operating data, each of the missing operating data is obtained by deleting the complete operating data corresponding to the missing operating data;
  • the network parameters of the target prediction network are updated according to the predicted fault operation data and the complete operation data corresponding to the missing operation data, and it is determined whether the updated target prediction network reaches the training target. If not, the step of inputting one of the missing operation data in the second training data set into the untrained target prediction network is continued until the updated target prediction network reaches the training target, thereby obtaining the trained target prediction network.
  • the data amplification is performed according to each of the restored fault operation data to obtain a plurality of amplified fault operation data, including:
  • each of the restored fault operation data and each of the data styles a number of the amplified fault operation data are determined, wherein the several amplified fault operation data respectively correspond to different combinations of the restored fault operation data and the data styles, and each of the amplified fault operation data is obtained based on the fusion of the restored fault operation data and the data style corresponding to the amplified fault operation data.
  • determining a first training data set corresponding to the fault prediction model according to the plurality of augmented fault operation data includes:
  • the first training data set is determined according to each of the amplified fault operation data, each of the basic fault operation data and each of the standard operation data, wherein each of the amplified fault operation data and each of the basic fault operation data corresponds to a first classification label, respectively, and each of the standard operation data corresponds to a second classification label, respectively.
  • the method further comprises:
  • an embodiment of the present invention further provides a device for generating a fault prediction model for an energy storage device in an extreme state, wherein the device comprises:
  • An acquisition module used to acquire a number of extreme fault operation data corresponding to a target device, wherein the extreme fault operation data is used to reflect the operation data of the target device when a fault occurs in an extreme environment, and the extreme environment is an environment in which a collection frequency is lower than a preset threshold;
  • a restoration module used for restoring each of the extreme fault operation data to obtain restored fault operation data corresponding to each of the extreme fault operation data
  • an amplification module used for performing data amplification according to each of the restored fault operation data to obtain a number of amplified fault operation data
  • a determination module configured to obtain a fault prediction model corresponding to the target device, and determine a first training data set corresponding to the fault prediction model according to a number of the amplified fault operation data
  • a training module is used to train the fault prediction model according to the first training data set, and determine a target fault prediction model corresponding to the target device according to the trained fault prediction model.
  • an embodiment of the present invention further provides a terminal, wherein the terminal includes a memory and one or more processors; the memory stores one or more programs; the program includes instructions for executing any of the above-described methods for generating a fault prediction model for an energy storage device applicable to extreme conditions; and the processor is used to execute the program.
  • an embodiment of the present invention further provides a computer-readable storage medium having a plurality of instructions stored thereon, wherein the instructions are suitable for being loaded and executed by a processor to implement any of the steps of the above-mentioned method for generating a fault prediction model for an energy storage device in extreme conditions.
  • the embodiments of the present invention restore and amplify the extreme fault operation data of the target device, thereby increasing the fault operation data of the target device in extreme environments, so that sufficient extreme fault operation data can be used to train the fault prediction model of the target device. This solves the problem in the prior art that the existing fault prediction model is difficult to apply to the fault prediction of energy storage devices in extreme environments due to the lack of sufficient historical fault data of energy storage devices in extreme environments.
  • FIG1 is a schematic flow chart of a method for generating a fault prediction model for an energy storage device in extreme conditions provided in an embodiment of the present invention.
  • FIG2 is a schematic diagram of the internal modules of a device for generating a fault prediction model for an energy storage device in an extreme state provided by an embodiment of the present invention.
  • FIG3 is a functional block diagram of a terminal provided by an embodiment of the present invention.
  • the present invention discloses a method for generating a fault prediction model for an energy storage device in an extreme state.
  • the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
  • the present invention provides a method for generating a fault prediction model for an energy storage device in extreme conditions, the method comprising: obtaining a number of extreme fault operation data corresponding to a target device, wherein the extreme fault operation data is used to reflect the operation data of the target device when a fault occurs in an extreme environment, and the extreme environment is an environment in which the acquisition frequency is lower than a preset threshold; restoring each of the extreme fault operation data to obtain restored fault operation data corresponding to each of the extreme fault operation data; performing data amplification according to each of the restored fault operation data to obtain a number of amplified fault operation data; obtaining a fault prediction model corresponding to the target device, and determining a first training data set corresponding to the fault prediction model according to the amplified fault operation data; training the fault prediction model according to the first training data set, and determining a target fault prediction model corresponding to the target device according to the trained fault prediction model.
  • the present invention increases the fault operation data of the target device in an extreme environment by restoring and amplifying the extreme fault operation data of the target device, so that a sufficient amount of extreme fault operation data can be used to train the fault prediction model of the target device.
  • the present invention solves the problem that the existing fault prediction model is difficult to apply to the fault prediction of energy storage devices in extreme environments due to the lack of sufficient historical fault data of energy storage devices in extreme environments.
  • the method comprises the following steps:
  • Step S100 Acquire a number of extreme fault operation data corresponding to the target device, wherein the extreme fault operation data is used to reflect the operation data of the target device when a fault occurs in an extreme environment, and the extreme environment is an environment where the collection frequency is lower than a preset threshold.
  • the target device in this embodiment can be any energy storage device that needs to be monitored for faults.
  • this embodiment needs to first obtain the fault operation data of the target device in extreme environments in the historical data, that is, to obtain extreme fault operation data.
  • an extreme environment refers to an environment where the collection rate is lower than a preset threshold.
  • the number of collections in environment A is 1, the number of collections in environment B is 10, the number of collections in environment C is 2, and the number of collections in environment D is 7.
  • the preset threshold is 15%, and environments A and C are extreme environments.
  • each extreme fault operation data is voltage data collected when the target device fails in an extreme state
  • each extreme fault operation data is current data collected when the target device fails in an extreme state.
  • the method further comprises the following steps:
  • Step S200 restore each of the extreme fault operation data to obtain restored fault operation data corresponding to each of the extreme fault operation data.
  • this embodiment needs to first restore each extreme fault operation data to repair the missing data in each extreme fault operation data, that is, to obtain multiple restored fault operation data.
  • each of the extreme fault operation data is time series data
  • each of the time series data includes a plurality of elements
  • the plurality of elements are respectively used to reflect the operation values corresponding to the same operation parameter at different time points.
  • the step S200 specifically includes the following steps:
  • Step S201 determining whether each of the time series data has missing elements, and using the time series data with the missing elements as the sequence data to be restored;
  • Step S202 restore each of the to-be-restored sequence data to obtain the restored fault operation data corresponding to each of the time series data.
  • the extreme fault operation data in this embodiment is in the form of time series data, each of which includes multiple elements, and each element has a time tag for indicating the collection time point corresponding to the data, so each element can reflect the operation value of the same operation parameter at different time points.
  • each element in time series data A can reflect the current value of the target device at different time points.
  • this embodiment needs to first filter out the time series data with missing elements, that is, to obtain the sequence data to be restored. Then, restore each time series data to be restored, and fill in the missing elements in each restored time series data, that is, to obtain the restored fault operation data.
  • step S202 specifically includes the following steps:
  • Step S2021 inputting each of the to-be-restored fault operation data into a target prediction network, wherein the target prediction network is pre-trained;
  • Step S2022 Obtain the restored fault operation data corresponding to each of the fault operation data to be restored output by the target prediction network based on each of the fault operation data to be restored.
  • this embodiment pre-constructs a target prediction network, which pre-learns the mapping relationship between each missing operating data and the corresponding complete operating data. Therefore, after each fault operating data to be restored is input into the target prediction network, the complete operating data corresponding to each fault operating data to be restored can be predicted, thereby completing the data recovery process.
  • the training process of the target prediction network includes:
  • Step S20211 obtaining a second training data set, wherein the second training data set includes a number of missing operating data and a number of complete operating data corresponding to the missing operating data, each of the missing operating data is obtained by deleting the complete operating data corresponding to the missing operating data;
  • Step S20212 inputting one of the missing operation data in the second training data set into the target prediction network that has not been trained, to obtain predicted fault operation data corresponding to the missing operation data;
  • Step S20213 updating the network parameters of the target prediction network according to the predicted fault operation data and the complete operation data corresponding to the missing operation data, and determining whether the updated target prediction network reaches the training target; if not, continuing to execute the step of inputting one of the missing operation data in the second training data set into the untrained target prediction network until the updated target prediction network reaches the training target, thereby obtaining the trained target prediction network.
  • each complete operation data in this embodiment is the operation data with no missing values collected in the early stage of a failure of the target device in an extreme environment; each missing operation data is obtained by deleting each complete operation data, and is used to reflect the operation data with missing values collected in the late stage of a failure of the target device in an extreme environment.
  • each missing operation data is obtained by deleting each complete operation data, and is used to reflect the operation data with missing values collected in the late stage of a failure of the target device in an extreme environment.
  • a reverse prediction network is obtained, wherein the input-output mapping relationship corresponding to the reverse prediction network and the target prediction network is a mutually inverse relationship, that is, the input of the reverse prediction network is the complete operation data, and the output is the missing operation data, and the initial network parameters corresponding to the reverse prediction network are determined based on the first network parameters corresponding to the target prediction network and the mutually inverse relationship;
  • the first network parameters are updated according to the second network parameters and the inverse relationship to obtain the target prediction network that has been updated again.
  • this embodiment uses the first network parameters corresponding to the trained target prediction network to construct the initial network parameters of the reverse prediction network, and then adjusts the second training data set corresponding to the target prediction network to a third training data set suitable for the reverse prediction network, and then uses the third training data set to update the network parameters of the reverse prediction network.
  • the first network parameters corresponding to the target prediction network are updated according to the second network parameters corresponding to the trained reverse prediction network, thereby further improving the prediction performance of the target prediction network.
  • the method further comprises the following steps:
  • Step S300 performing data amplification according to each of the restored fault operation data to obtain a plurality of amplified fault operation data.
  • this embodiment needs to perform data amplification based on each restored fault operation data, thereby increasing the fault operation data under extreme environments, that is, obtaining multiple amplified fault operation data.
  • step S300 specifically includes the following steps:
  • Step S301 acquiring data styles corresponding to each of the restored fault operation data, and obtaining a plurality of data styles, wherein the data style corresponding to each of the restored fault operation data is used to reflect the data change characteristics corresponding to the restored fault operation data;
  • Step S302 determine a number of the amplified fault operation data according to each of the restored fault operation data and each of the data styles, wherein the several amplified fault operation data respectively correspond to different combinations of the restored fault operation data and the data styles, and each of the amplified fault operation data is obtained by fusion of the restored fault operation data and the data style corresponding to the amplified fault operation data.
  • the data styles of fault operation data collected in different types of extreme environments will also be different.
  • the data change characteristics of the fault operation data collected in extreme environment A are a sudden drop over time
  • the data change characteristics of the fault operation data collected in extreme environment B are a sudden increase over time. Therefore, in order to expand the fault operation data in extreme environments, this embodiment requires that each restored fault operation data learns each other's data styles, so as to combine more fault operation data in extreme environments.
  • the method further comprises the following steps:
  • Step S400 Acquire a fault prediction model corresponding to the target device, and determine a first training data set corresponding to the fault prediction model according to a number of the amplified fault operation data.
  • this embodiment needs to generate training data corresponding to the fault prediction model according to each augmented fault operation data, that is, to obtain a first training data set. Since each augmented fault operation data already contains a sufficient amount of fault operation data under extreme environments, training the fault prediction model through the first training data set can enable it to fully learn the data characteristics of the fault operation data under extreme environments, thereby improving its prediction performance for the operation data collected under extreme environments.
  • step S400 specifically includes the following steps:
  • Step S401 Acquire a number of basic fault operation data corresponding to the target device, wherein the basic fault operation data is used to reflect the operation data of the target device when a fault occurs in a non-extreme environment;
  • Step S402 Acquire a number of standard operation data corresponding to the target device, wherein the standard operation data is used to reflect the operation data of the target device when no failure occurs;
  • Step S403 determine the first training data set according to each of the amplified fault operation data, each of the basic fault operation data, and each of the standard operation data, wherein each of the amplified fault operation data and each of the basic fault operation data corresponds to a first classification label, respectively, and each of the standard operation data corresponds to a second classification label, respectively.
  • the present embodiment in order to fully learn the data characteristics of the operating data of the target device in a fault state, the present embodiment also needs to obtain the fault operating data of the target device in a non-extreme environment, that is, to obtain basic fault operating data. At the same time, in order to learn the data characteristics of the operating data of the target device in a normal state, the present embodiment also needs to collect the operating data of the target device when no fault occurs, that is, to obtain standard operating data. Finally, a first training data set corresponding to the fault prediction model is constructed based on all the amplified fault operating data, basic fault operating data, and standard operating data.
  • the fault prediction model can fully learn the difference between the standard operating data and the fault operating data based on the first training data set, thereby improving the classification performance of the fault prediction model.
  • the method further comprises the following steps:
  • Step S500 training the fault prediction model according to the first training data set, and determining a target fault prediction model corresponding to the target device according to the trained fault prediction model.
  • the first training data set contains sufficient operating data of the target device when a failure occurs in an extreme environment
  • using the first training data set to train the fault prediction model can improve the accuracy of the fault prediction model in classifying the operating data collected in the extreme environment.
  • the method further includes the following steps:
  • Step S600 obtaining current operation data corresponding to the target device, and inputting the current operation data into the target fault prediction model;
  • Step S601 Obtain the current operating state corresponding to the target device output by the target fault prediction model based on the current operating data.
  • the target fault prediction model is a model trained based on the first training data set, it can not only accurately classify the operating data of the target device collected in a common environment, but also accurately classify the operating data of the target device collected in an extreme environment, thereby accurately judging the current operating status of the target device based on the current operating data of the target device.
  • the present invention further provides a device for generating a fault prediction model for an energy storage device in an extreme state, as shown in FIG2 , the device comprising:
  • Acquisition module 01 used to acquire a number of extreme fault operation data corresponding to the target device, wherein the extreme fault operation data is used to reflect the operation data of the target device when a fault occurs in an extreme environment, and the extreme environment is an environment where the acquisition frequency is lower than a preset threshold;
  • a restoration module 02 used for restoring each of the extreme fault operation data to obtain restored fault operation data corresponding to each of the extreme fault operation data
  • An amplification module 03 is used to amplify data according to each of the restored fault operation data to obtain a number of amplified fault operation data;
  • a determination module 04 is used to obtain a fault prediction model corresponding to the target device, and determine a first training data set corresponding to the fault prediction model according to a number of the amplified fault operation data;
  • the training module 05 is used to train the fault prediction model according to the first training data set, and determine the target fault prediction model corresponding to the target device according to the trained fault prediction model.
  • the present invention also provides a terminal, whose principle block diagram can be shown in Figure 3.
  • the terminal includes a processor, a memory, a network interface, and a display screen connected via a system bus.
  • the processor of the terminal is used to provide computing and control capabilities.
  • the memory of the terminal includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system and a computer program.
  • the internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium.
  • the network interface of the terminal is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a fault prediction model generation method for energy storage devices suitable for extreme states is implemented.
  • the display screen of the terminal can be a liquid crystal display or an electronic ink display.
  • FIG3 is only a block diagram of a partial structure related to the solution of the present invention, and does not constitute a limitation on the terminal to which the solution of the present invention is applied.
  • the specific terminal may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.
  • one or more programs are stored in the memory of the terminal, and are configured to be executed by one or more processors.
  • the one or more programs include instructions for performing a method for generating a fault prediction model for an energy storage device applicable to extreme conditions.
  • Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
  • the present invention discloses a method for generating a fault prediction model for an energy storage device in extreme conditions, the method comprising: obtaining a number of extreme fault operation data corresponding to a target device, wherein the extreme fault operation data is used to reflect the operation data of the target device when a fault occurs in an extreme environment, and the extreme environment is an environment in which the acquisition frequency is lower than a preset threshold; restoring each of the extreme fault operation data to obtain restored fault operation data corresponding to each of the extreme fault operation data; performing data amplification according to each of the restored fault operation data to obtain a number of amplified fault operation data; obtaining a fault prediction model corresponding to the target device, and determining a first training data set corresponding to the fault prediction model according to the amplified fault operation data; training the fault prediction model according to the first training data set, and determining a target fault prediction model corresponding to the target device according to the trained fault prediction model.
  • the present invention increases the fault operation data of the target device in an extreme environment by restoring and amplifying the extreme fault operation data of the target device, so that a sufficient amount of extreme fault operation data can be used to train the fault prediction model of the target device.
  • the present invention solves the problem that the existing fault prediction model is difficult to apply to the fault prediction of energy storage devices in extreme environments due to the lack of sufficient historical fault data of energy storage devices in extreme environments.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明公开了一种适用于极端状态的储能设备的故障预测模型生成方法，本发明通过对目标设备的极端故障运行数据进行复原、扩增，增多了目标设备在极端环境下的故障运行数据，因此可以采用足量的极端故障运行数据对目标设备的故障预测模型进行训练。解决了现有技术中由于缺乏足量的储能设备在极端环境中的历史故障数据，因此现有的故障预测模型难以适用于极端环境中的储能设备的故障预测问题。

Description

一种适用于极端状态的储能设备的故障预测模型生成方法 技术领域
本发明涉及模型生成领域，尤其涉及的是一种适用于极端状态的储能设备的故障预测模型生成方法。
 背景技术
传统的储能设备的故障监测主要依赖于专业的工作人员实时监控储能设备的运行数据，需要耗费大量的人力。并且由于人工监测具有一定的主观性，当工作人员的经验不足或者对当前储能设备不够了解时，就难以准确预测出储能设备的故障，从而影响储能系统的正常运作。随着科学技术的不断进步，基于机器学习的故障预测模型已经逐渐代替了人工监测。故障预测模型是通过学习历史参数的特性，来判断未来参数所反映的设备状态。因此故障预测模型需要大量的历史数据进行训练。然而储能设备在极端环境中的历史故障数据获取难度大，且获取数量少，例如雷击，高温，高压，极寒，真空等极端户外或太空环境中获取储能设备的历史故障数据是机器困难的。由于缺乏足量的储能设备在极端环境中的历史故障数据，因此现有的故障预测模型难以适用于极端环境中的储能设备的故障预测。
因此，现有技术还有待改进和发展。
 技术问题
本发明要解决的技术问题在于，针对现有技术的上述缺陷，提供一种适用于极端状态的储能设备的故障预测模型生成方法，旨在解决现有技术中由于缺乏足量的储能设备在极端环境中的历史故障数据，因此现有的故障预测模型难以适用于极端环境中的储能设备的故障预测问题。
 技术解决方案
本发明解决问题所采用的技术方案如下：
第一方面，本发明实施例提供一种适用于极端状态的储能设备的故障预测模型生成方法，其中，所述方法包括：
获取目标设备对应的若干极端故障运行数据，其中，所述极端故障运行数据用于反映所述目标设备在极端环境内发生故障时的运行数据，所述极端环境为采集频率低于预设阈值的环境；
对各所述极端故障运行数据进行复原，得到各所述极端故障运行数据分别对应的复原故障运行数据；
根据各所述复原故障运行数据进行数据扩增，得到若干扩增故障运行数据；
获取所述目标设备对应的故障预测模型，根据若干所述扩增故障运行数据确定所述故障预测模型对应的第一训练数据集；
根据所述第一训练数据集对所述故障预测模型进行训练，根据已训练的所述故障预测模型确定所述目标设备对应的目标故障预测模型。
在一种实施方式中，各所述极端故障运行数据均为时间序列数据，每一所述时间序列数据包括若干元素，若干所述元素分别用于反映同一运行参数在不同时间点分别对应的运行值，所述对各所述极端故障运行数据进行复原，得到各所述极端故障运行数据分别对应的复原故障运行数据，包括：
判断各所述时间序列数据是否存在缺失元素，将存在所述缺失元素的所述时间序列数据作为待复原序列数据；
对各所述待复原序列数据进行复原，得到各所述时间序列数据分别对应的所述复原故障运行数据。
在一种实施方式中，所述对各所述待复原序列数据进行复原，得到各所述时间序列数据分别对应的所述复原故障运行数据，包括：
将各所述待复原故障运行数据输入目标预测网络，其中，所述目标预测网络预先经过训练；
获取所述目标预测网络基于各所述待复原故障运行数据输出的各所述待复原故障运行数据分别对应的所述复原故障运行数据。
在一种实施方式中，所述目标预测网络的训练过程包括：
获取第二训练数据集，其中，所述第二训练数据集包括若干缺失运行数据和若干所述缺失运行数据分别对应的完整运行数据，每一所述缺失运行数据基于对该缺失运行数据对应的所述完整运行数据进行删减得到；
将所述第二训练数据集中的一个所述缺失运行数据输入未训练完毕的所述目标预测网络，得到该缺失运行数据对应的预测故障运行数据；
根据所述预测故障运行数据和该缺失运行数据对应的所述完整运行数据对所述目标预测网络的网络参数进行更新，判断更新后的所述目标预测网络是否达到训练目标，若否，继续执行将所述第二训练数据集中的一个所述缺失运行数据输入未训练完毕的所述目标预测网络的步骤，直至更新后的所述目标预测网络达到所述训练目标，得到已训练完毕的所述目标预测网络。
在一种实施方式中，所述根据各所述复原故障运行数据进行数据扩增，得到若干扩增故障运行数据，包括：
获取各所述复原故障运行数据分别对应的数据风格，得到若干所述数据风格，其中，每一所述复原故障运行数据对应的所述数据风格用于反映该复原故障运行数据对应的数据变化特征；
根据各所述复原故障运行数据和各所述数据风格，确定若干所述扩增故障运行数据，其中，若干所述扩增故障运行数据分别对应不同的所述复原故障运行数据和所述数据风格的组合，每一所述扩增故障运行数据基于该扩增故障运行数据对应的所述复原故障运行数据和所述数据风格融合得到。
在一种实施方式中，所述根据若干所述扩增故障运行数据确定所述故障预测模型对应的第一训练数据集，包括：
获取所述目标设备对应的若干基础故障运行数据，其中，所述基础故障运行数据用于反映所述目标设备在非极端环境内发生故障时的运行数据；
获取所述目标设备对应的若干标准运行数据，其中，所述标准运行数据用于反映所述目标设备在未发生故障时的运行数据；
根据各所述扩增故障运行数据、各所述基础故障运行数据以及各所述标准运行数据，确定所述第一训练数据集，其中，各所述扩增故障运行数据和各所述基础故障运行数据分别对应第一分类标签，各所述标准运行数据分别对应第二分类标签。
在一种实施方式中，所述方法还包括：
获取所述目标设备对应的当前运行数据，将所述当前运行数据输入所述目标故障预测模型；
获取所述目标故障预测模型基于所述当前运行数据输出的所述目标设备对应的当前运行状态。
第二方面，本发明实施例还提供一种适用于极端状态的储能设备的故障预测模型生成装置，其中，所述装置包括：
获取模块，用于获取目标设备对应的若干极端故障运行数据，其中，所述极端故障运行数据用于反映所述目标设备在极端环境内发生故障时的运行数据，所述极端环境为采集频率低于预设阈值的环境；
复原模块，用于对各所述极端故障运行数据进行复原，得到各所述极端故障运行数据分别对应的复原故障运行数据；
扩增模块，用于根据各所述复原故障运行数据进行数据扩增，得到若干扩增故障运行数据；
确定模块，用于获取所述目标设备对应的故障预测模型，根据若干所述扩增故障运行数据确定所述故障预测模型对应的第一训练数据集；
训练模块，用于根据所述第一训练数据集对所述故障预测模型进行训练，根据已训练的所述故障预测模型确定所述目标设备对应的目标故障预测模型。
第三方面，本发明实施例还提供一种终端，其中，所述终端包括有存储器和一个或者一个以上处理器；所述存储器存储有一个或者一个以上的程序；所述程序包含用于执行如上述任一所述的适用于极端状态的储能设备的故障预测模型生成方法的指令；所述处理器用于执行所述程序。
第四方面，本发明实施例还提供一种计算机可读存储介质，其上存储有多条指令，其中，所述指令适用于由处理器加载并执行，以实现上述任一所述的适用于极端状态的储能设备的故障预测模型生成方法的步骤。
 有益效果
本发明的有益效果：本发明实施例通过对目标设备的极端故障运行数据进行复原、扩增，增多了目标设备在极端环境下的故障运行数据，因此可以采用足量的极端故障运行数据对目标设备的故障预测模型进行训练。解决了现有技术中由于缺乏足量的储能设备在极端环境中的历史故障数据，因此现有的故障预测模型难以适用于极端环境中的储能设备的故障预测问题。
 附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明中记载的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本发明实施例提供的适用于极端状态的储能设备的故障预测模型生成方法的流程示意图。
图2是本发明实施例提供的适用于极端状态的储能设备的故障预测模型生成装置的内部模块示意图。
图3是本发明实施例提供的终端的原理框图。
 本发明的实施方式
本发明公开了一种适用于极端状态的储能设备的故障预测模型生成方法，为使本发明的目的、技术方案及效果更加清楚、明确，以下参照附图并举实施例对本发明进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本发明，并不用于限定本发明。
本技术领域技术人员可以理解，除非特意声明，这里使用的单数形式“一”、“一个”、“所述”和“该”也可包括复数形式。应该进一步理解的是，本发明的说明书中使用的措辞“包括”是指存在所述特征、整数、步骤、操作、元件和/或组件，但是并不排除存在或添加一个或多个其他特征、整数、步骤、操作、元件、组件和/或它们的组。 应该理解，当我们称元件被“连接”或“耦接”到另一元件时，它可以直接连接或耦接到其他元件，或者也可以存在中间元件。此外，这里使用的“连接”或“耦接”可以包括无线连接或无线耦接。这里使用的措辞“和/或”包括一个或更多个相关联的列出项的全部或任一单元和全部组合。
本技术领域技术人员可以理解，除非另外定义，这里使用的所有术语(包括技术术语和科学术语)，具有与本发明所属领域中的普通技术人员的一般理解相同的意义。还应该理解的是，诸如通用字典中定义的那些术语，应该被理解为具有与现有技术的上下文中的意义一致的意义，并且除非像这里一样被特定定义，否则不会用理想化或过于正式的含义来解释。
针对现有技术的上述缺陷，本发明提供一种适用于极端状态的储能设备的故障预测模型生成方法，所述方法包括：获取目标设备对应的若干极端故障运行数据，其中，所述极端故障运行数据用于反映所述目标设备在极端环境内发生故障时的运行数据，所述极端环境为采集频率低于预设阈值的环境；对各所述极端故障运行数据进行复原，得到各所述极端故障运行数据分别对应的复原故障运行数据；根据各所述复原故障运行数据进行数据扩增，得到若干扩增故障运行数据；获取所述目标设备对应的故障预测模型，根据若干所述扩增故障运行数据确定所述故障预测模型对应的第一训练数据集；根据所述第一训练数据集对所述故障预测模型进行训练，根据已训练的所述故障预测模型确定所述目标设备对应的目标故障预测模型。本发明通过对目标设备的极端故障运行数据进行复原、扩增，增多了目标设备在极端环境下的故障运行数据，因此可以采用足量的极端故障运行数据对目标设备的故障预测模型进行训练。解决了现有技术中由于缺乏足量的储能设备在极端环境中的历史故障数据，因此现有的故障预测模型难以适用于极端环境中的储能设备的故障预测问题。
如图1所示，所述方法包括如下步骤：
步骤S100、获取目标设备对应的若干极端故障运行数据，其中，所述极端故障运行数据用于反映所述目标设备在极端环境内发生故障时的运行数据，所述极端环境为采集频率低于预设阈值的环境。
具体地，本实施例中的目标设备可以是任意一台需要进行故障监测的储能设备，为了使得最终生成模型可以适用于极端环境下的设备故障预测，因此本实施例需要首先获得历史数据中目标设备在极端环境下的故障运行数据，即得到极端故障运行数据。需要说明的是，极端环境指的是采集评率低于预设阈值的环境，例如环境A的采集次数为1，环境B的采集次数为10，环境C的采集次数为2，环境D的采集次数为7，则环境A的采集频率为1/20=5%，环境B的采集频率为10/20=50%，环境C的采集频率为2/20=10%，环境D的采集频率为7/20=35%，预设阈值为15%，则环境A、C为极端环境。
需要说明的是，若干所述极端故障运行数据分别对应的数据类型相同。例如，各极端故障运行数据均为目标设备在极端状态下发生故障时采集到的电压数据，或者各极端故障运行数据均为目标设备在极端状态下发生故障时采集到的电流数据。
如图1所示，所述方法还包括如下步骤：
步骤S200、对各所述极端故障运行数据进行复原，得到各所述极端故障运行数据分别对应的复原故障运行数据。
具体地，由于极端状态下获取的故障运行数据可能存在数据缺失，例如缺失采集时间段内的个别时间点的运行数据，因此为了保证数据的有效性和可靠性，本实施例需要先对各极端故障运行数据进行复原，以修复各极端故障运行数据中的缺失数据，即得到多个复原故障运行数据。
在一种实现方式中，各所述极端故障运行数据均为时间序列数据，每一所述时间序列数据包括若干元素，若干所述元素分别用于反映同一运行参数在不同时间点分别对应的运行值，所述步骤S200具体包括如下步骤：
步骤S201、判断各所述时间序列数据是否存在缺失元素，将存在所述缺失元素的所述时间序列数据作为待复原序列数据；
步骤S202、对各所述待复原序列数据进行复原，得到各所述时间序列数据分别对应的所述复原故障运行数据。
具体地，本实施例中的极端故障运行数据为时间序列数据的形式，每一时间序列数据都包括多个元素，每一元素都具有时间标签用于指示该数据对应的采集时间点，因此各元素可以反映同一运行参数在不同时间点的运行值，例如时间序列数据A中各元素可以反映目标设备在不同时间点的电流值。为了保障各时间序列数据的有效性和可靠性，本实施例需要先筛选出缺失元素的时间序列数据，即得到待复原序列数据。然后针对每一待复原时间序列数据进行复原，将各复原时间序列数据中的缺失元素补齐，即得到复原故障运行数据。
在一种实现方式中，所述步骤S202具体包括如下步骤：
步骤S2021、将各所述待复原故障运行数据输入目标预测网络，其中，所述目标预测网络预先经过训练；
步骤S2022、获取所述目标预测网络基于各所述待复原故障运行数据输出的各所述待复原故障运行数据分别对应的所述复原故障运行数据。
具体地，本实施例预先构建了一个目标预测网络，该目标预测网络预先学习了各缺失运行数据与对应的完整运行数据之间的映射关系，因此将各待复原故障运行数据输入目标预测网络以后，即可预测出各待复原故障运行数据分别对应的完整运行数据，从而完成数据复原过程。
在一种实现方式中，所述目标预测网络的训练过程包括：
步骤S20211、获取第二训练数据集，其中，所述第二训练数据集包括若干缺失运行数据和若干所述缺失运行数据分别对应的完整运行数据，每一所述缺失运行数据基于对该缺失运行数据对应的所述完整运行数据进行删减得到；
步骤S20212、将所述第二训练数据集中的一个所述缺失运行数据输入未训练完毕的所述目标预测网络，得到该缺失运行数据对应的预测故障运行数据；
步骤S20213、根据所述预测故障运行数据和该缺失运行数据对应的所述完整运行数据对所述目标预测网络的网络参数进行更新，判断更新后的所述目标预测网络是否达到训练目标，若否，继续执行将所述第二训练数据集中的一个所述缺失运行数据输入未训练完毕的所述目标预测网络的步骤，直至更新后的所述目标预测网络达到所述训练目标，得到已训练完毕的所述目标预测网络。
具体地，本实施例中的各完整运行数据是目标设备在极端环境中发生故障早期采集到的未缺失数值的运行数据；各缺失运行数据则是根据对各完整运行数据进行删减得到的，用于反映目标设备在极端环境中发生故障晚期采集到的缺失数值的运行数据。根据将各缺失运行数据作为训练数据，将各完整运行数据作为真实标签，以构建目标预测网络对应的第二训练数据集，通过第二训练数据集对目标预测网络进行训练，可以使其充分学习到缺失运行数据与对应的完整运行数据之间的映射关系，从而提高目标预测网络的预测性能。
在一种实现方式中，获取逆向预测网络，其中，所述逆向预测网络与所述目标预测网络分别对应的输入输出映射关系为互逆关系，即所述逆向预测网络的输入为所述完整运行数据，输出为所述缺失运行数据，所述逆向预测网络对应的初始网络参数基于所述目标预测网络对应的第一网络参数和所述互逆关系确定；
根据所述第二训练数据集构建第三训练数据集，其中，所述第三训练数据集中各所述缺失运行数据为训练数据，各所述缺失运行数据分别对应的所述完整运行数据为真实标签；
根据所述第三训练数据集对所述逆向预测网络进行迭代更新，直至达到所述逆向预测网络对应的训练目标，得到已训练的所述逆向预测网络；
获取已训练的所述逆向预测网络对应的第二网络参数；
根据所述第二网络参数和所述互逆关系对所述第一网络参数进行更新，得到再次更新后的所述目标预测网络。
具体地，由于目标预测网络和逆向预测网络之间的输入输出映射关系为互逆关系，因此两者分别对应的网络参数之间具有一定的关联，所以本实施例采用已训练的目标预测网络对应的第一网络参数构建逆向预测网络的初始网络参数，然后将目标预测网络对应的第二训练数据集调整为适用于逆向预测网络的第三训练数据集，然后采用第三训练数据集对逆向预测网络的网络参数进行更新，当逆向预测网络训练完毕以后，再根据训练后的逆向预测网络对应的第二网络参数对目标预测网络对应的第一网络参数进行更新，从而进一步提升目标预测网络的预测性能。
如图1所示，所述方法还包括如下步骤：
步骤S300、根据各所述复原故障运行数据进行数据扩增，得到若干扩增故障运行数据。
具体地，由于极端环境下的故障运行数据数量过少，难以直接应用于模型训练过程。因此本实施例需要根据各个复原故障运行数据进行数据扩增，从而增多极端环境下的故障运行数据，即得到多个扩增故障运行数据。
在一种实现方式中，所述步骤S300具体包括如下步骤：
步骤S301、获取各所述复原故障运行数据分别对应的数据风格，得到若干所述数据风格，其中，每一所述复原故障运行数据对应的所述数据风格用于反映该复原故障运行数据对应的数据变化特征；
步骤S302、根据各所述复原故障运行数据和各所述数据风格，确定若干所述扩增故障运行数据，其中，若干所述扩增故障运行数据分别对应不同的所述复原故障运行数据和所述数据风格的组合，每一所述扩增故障运行数据基于该扩增故障运行数据对应的所述复原故障运行数据和所述数据风格融合得到。
具体地，由于极端环境也分为多种不同类型，不同类型的极端环境下采集的故障运行数据的数据风格也会有所差异，例如在极端环境A下采集到的故障运行数据的数据变化特征是随着时间骤降，在极端环境B下采集到的故障运行数据的数据变化特征是随着时间骤升。因此为了扩增极端环境下的故障运行数据，本实施例需要各复原故障运行数据互相学习彼此的数据风格，从而组合出更多的极端环境下的故障运行数据。
如图1所示，所述方法还包括如下步骤：
步骤S400、获取所述目标设备对应的故障预测模型，根据若干所述扩增故障运行数据确定所述故障预测模型对应的第一训练数据集。
具体地，为了提高故障预测模型对极端环境下采集到的运行数据的预测性能，本实施例需要根据各扩增故障运行数据生成故障预测模型对应的训练数据，即得到第一训练数据集。由于各扩增故障运行数据已经包含足量的极端环境下的故障运行数据，因此通过第一训练数据集对故障预测模型进行训练可以使其充分学习到极端环境下的故障运行数据的数据特征，从而提高其对极端环境下采集到的运行数据的预测性能。
在一种实现方式中，所述步骤S400具体包括如下步骤：
步骤S401、获取所述目标设备对应的若干基础故障运行数据，其中，所述基础故障运行数据用于反映所述目标设备在非极端环境内发生故障时的运行数据；
步骤S402、获取所述目标设备对应的若干标准运行数据，其中，所述标准运行数据用于反映所述目标设备在未发生故障时的运行数据；
步骤S403、根据各所述扩增故障运行数据、各所述基础故障运行数据以及各所述标准运行数据，确定所述第一训练数据集，其中，各所述扩增故障运行数据和各所述基础故障运行数据分别对应第一分类标签，各所述标准运行数据分别对应第二分类标签。
具体地，为了充分学习目标设备在故障状态时的运行数据的数据特征，本实施例还需要获取目标设备在非极端环境下的故障运行数据，即得到基础故障运行数据。同时，为了学习目标设备在正常状态时的运行数据的数据特征，本实施例还需要采集目标设备在未发生故障时的运行数据，即得到标准运行数据。最后根据所有扩增故障运行数据、基础故障运行数据以及标准运行数据构建出故障预测模型对应的第一训练数据集，由于扩增故障运行数据和基础故障运行数据对应的是第一分类标签，标准运行数据对应的是第二分类标签，因此故障预测模型基于第一训练数据集可以充分学习到标准运行数据和故障运行数据之间的区别，从而提高故障预测模型的分类性能。
如图1所示，所述方法还包括如下步骤：
步骤S500、根据所述第一训练数据集对所述故障预测模型进行训练，根据已训练的所述故障预测模型确定所述目标设备对应的目标故障预测模型。
具体地，由于第一训练数据集中包含有足量的目标设备在极端环境下发生故障时的运行数据，因此采用第一训练数据集训练故障预测模型，可以提升故障预测模型分类极端环境下采集到的运行数据的准确性。
在一种实现方式中，所述方法还包括如下步骤：
步骤S600、获取所述目标设备对应的当前运行数据，将所述当前运行数据输入所述目标故障预测模型；
步骤S601、获取所述目标故障预测模型基于所述当前运行数据输出的所述目标设备对应的当前运行状态。
具体地，由于目标故障预测模型是基于第一训练数据集训练得到的模型，因此其不仅能准确分类在常见环境中采集到的目标设备的运行数据，还能够准确分类在极端环境中采集到的目标设备的运行数据，从而准确地根据目标设备的当前运行数据判断出目标设备的当前运行状态。
基于上述实施例，本发明还提供了一种适用于极端状态的储能设备的故障预测模型生成装置，如图2所示，所述装置包括：
获取模块01，用于获取目标设备对应的若干极端故障运行数据，其中，所述极端故障运行数据用于反映所述目标设备在极端环境内发生故障时的运行数据，所述极端环境为采集频率低于预设阈值的环境；
复原模块02，用于对各所述极端故障运行数据进行复原，得到各所述极端故障运行数据分别对应的复原故障运行数据；
扩增模块03，用于根据各所述复原故障运行数据进行数据扩增，得到若干扩增故障运行数据；
确定模块04，用于获取所述目标设备对应的故障预测模型，根据若干所述扩增故障运行数据确定所述故障预测模型对应的第一训练数据集；
训练模块05，用于根据所述第一训练数据集对所述故障预测模型进行训练，根据已训练的所述故障预测模型确定所述目标设备对应的目标故障预测模型。
基于上述实施例，本发明还提供了一种终端，其原理框图可以如图3所示。该终端包括通过系统总线连接的处理器、存储器、网络接口、显示屏。其中，该终端的处理器用于提供计算和控制能力。该终端的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统和计算机程序。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该终端的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现适用于极端状态的储能设备的故障预测模型生成方法。该终端的显示屏可以是液晶显示屏或者电子墨水显示屏。
本领域技术人员可以理解，图3中示出的原理框图，仅仅是与本发明方案相关的部分结构的框图，并不构成对本发明方案所应用于其上的终端的限定，具体的终端可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一种实现方式中，所述终端的存储器中存储有一个或者一个以上的程序，且经配置以由一个或者一个以上处理器执行所述一个或者一个以上程序包含用于进行适用于极端状态的储能设备的故障预测模型生成方法的指令。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本发明所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器（ROM）、可编程ROM（PROM）、电可编程ROM（EPROM）、电可擦除可编程ROM（EEPROM）或闪存。易失性存储器可包括随机存取存储器（RAM）或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM（SRAM）、动态RAM（DRAM）、同步DRAM（SDRAM）、双数据率SDRAM（DDRSDRAM）、增强型SDRAM（ESDRAM）、同步链路（Synchlink） DRAM（SLDRAM）、存储器总线（Rambus）直接RAM（RDRAM）、直接存储器总线动态RAM（DRDRAM）、以及存储器总线动态RAM（RDRAM）等。
综上所述，本发明公开了一种适用于极端状态的储能设备的故障预测模型生成方法，所述方法包括：获取目标设备对应的若干极端故障运行数据，其中，所述极端故障运行数据用于反映所述目标设备在极端环境内发生故障时的运行数据，所述极端环境为采集频率低于预设阈值的环境；对各所述极端故障运行数据进行复原，得到各所述极端故障运行数据分别对应的复原故障运行数据；根据各所述复原故障运行数据进行数据扩增，得到若干扩增故障运行数据；获取所述目标设备对应的故障预测模型，根据若干所述扩增故障运行数据确定所述故障预测模型对应的第一训练数据集；根据所述第一训练数据集对所述故障预测模型进行训练，根据已训练的所述故障预测模型确定所述目标设备对应的目标故障预测模型。本发明通过对目标设备的极端故障运行数据进行复原、扩增，增多了目标设备在极端环境下的故障运行数据，因此可以采用足量的极端故障运行数据对目标设备的故障预测模型进行训练。解决了现有技术中由于缺乏足量的储能设备在极端环境中的历史故障数据，因此现有的故障预测模型难以适用于极端环境中的储能设备的故障预测问题。
应当理解的是，本发明的应用不限于上述的举例，对本领域普通技术人员来说，可以根据上述说明加以改进或变换，所有这些改进和变换都应属于本发明所附权利要求的保护范围。

Claims (10)

  1. 一种适用于极端状态的储能设备的故障预测模型生成方法，其特征在于，所述方法包括：
    获取目标设备对应的若干极端故障运行数据，其中，所述极端故障运行数据用于反映所述目标设备在极端环境内发生故障时的运行数据，所述极端环境为采集频率低于预设阈值的环境；
    对各所述极端故障运行数据进行复原，得到各所述极端故障运行数据分别对应的复原故障运行数据；
    根据各所述复原故障运行数据进行数据扩增，得到若干扩增故障运行数据；
    获取所述目标设备对应的故障预测模型，根据若干所述扩增故障运行数据确定所述故障预测模型对应的第一训练数据集；
    根据所述第一训练数据集对所述故障预测模型进行训练，根据已训练的所述故障预测模型确定所述目标设备对应的目标故障预测模型。
  2. 根据权利要求1所述的适用于极端状态的储能设备的故障预测模型生成方法，其特征在于，各所述极端故障运行数据均为时间序列数据，每一所述时间序列数据包括若干元素，若干所述元素分别用于反映同一运行参数在不同时间点分别对应的运行值，所述对各所述极端故障运行数据进行复原，得到各所述极端故障运行数据分别对应的复原故障运行数据，包括：
    判断各所述时间序列数据是否存在缺失元素，将存在所述缺失元素的所述时间序列数据作为待复原序列数据；
    对各所述待复原序列数据进行复原，得到各所述时间序列数据分别对应的所述复原故障运行数据。
  3. 根据权利要求2所述的适用于极端状态的储能设备的故障预测模型生成方法，其特征在于，所述对各所述待复原序列数据进行复原，得到各所述时间序列数据分别对应的所述复原故障运行数据，包括：
    将各所述待复原故障运行数据输入目标预测网络，其中，所述目标预测网络预先经过训练；
    获取所述目标预测网络基于各所述待复原故障运行数据输出的各所述待复原故障运行数据分别对应的所述复原故障运行数据。
  4. 根据权利要求3所述的适用于极端状态的储能设备的故障预测模型生成方法，其特征在于，所述目标预测网络的训练过程包括：
    获取第二训练数据集，其中，所述第二训练数据集包括若干缺失运行数据和若干所述缺失运行数据分别对应的完整运行数据，每一所述缺失运行数据基于对该缺失运行数据对应的所述完整运行数据进行删减得到；
    将所述第二训练数据集中的一个所述缺失运行数据输入未训练完毕的所述目标预测网络，得到该缺失运行数据对应的预测故障运行数据；
    根据所述预测故障运行数据和该缺失运行数据对应的所述完整运行数据对所述目标预测网络的网络参数进行更新，判断更新后的所述目标预测网络是否达到训练目标，若否，继续执行将所述第二训练数据集中的一个所述缺失运行数据输入未训练完毕的所述目标预测网络的步骤，直至更新后的所述目标预测网络达到所述训练目标，得到已训练完毕的所述目标预测网络。
  5. 根据权利要求1所述的适用于极端状态的储能设备的故障预测模型生成方法，其特征在于，所述根据各所述复原故障运行数据进行数据扩增，得到若干扩增故障运行数据，包括：
    获取各所述复原故障运行数据分别对应的数据风格，得到若干所述数据风格，其中，每一所述复原故障运行数据对应的所述数据风格用于反映该复原故障运行数据对应的数据变化特征；
    根据各所述复原故障运行数据和各所述数据风格，确定若干所述扩增故障运行数据，其中，若干所述扩增故障运行数据分别对应不同的所述复原故障运行数据和所述数据风格的组合，每一所述扩增故障运行数据基于该扩增故障运行数据对应的所述复原故障运行数据和所述数据风格融合得到。
  6. 根据权利要求1所述的适用于极端状态的储能设备的故障预测模型生成方法，其特征在于，所述根据若干所述扩增故障运行数据确定所述故障预测模型对应的第一训练数据集，包括：
    获取所述目标设备对应的若干基础故障运行数据，其中，所述基础故障运行数据用于反映所述目标设备在非极端环境内发生故障时的运行数据；
    获取所述目标设备对应的若干标准运行数据，其中，所述标准运行数据用于反映所述目标设备在未发生故障时的运行数据；
    根据各所述扩增故障运行数据、各所述基础故障运行数据以及各所述标准运行数据，确定所述第一训练数据集，其中，各所述扩增故障运行数据和各所述基础故障运行数据分别对应第一分类标签，各所述标准运行数据分别对应第二分类标签。
  7. 根据权利要求1所述的适用于极端状态的储能设备的故障预测模型生成方法，其特征在于，所述方法还包括：
    获取所述目标设备对应的当前运行数据，将所述当前运行数据输入所述目标故障预测模型；
    获取所述目标故障预测模型基于所述当前运行数据输出的所述目标设备对应的当前运行状态。
  8. 一种适用于极端状态的储能设备的故障预测模型生成装置，其特征在于，所述装置包括：
    获取模块，用于获取目标设备对应的若干极端故障运行数据，其中，所述极端故障运行数据用于反映所述目标设备在极端环境内发生故障时的运行数据，所述极端环境为采集频率低于预设阈值的环境；
    复原模块，用于对各所述极端故障运行数据进行复原，得到各所述极端故障运行数据分别对应的复原故障运行数据；
    扩增模块，用于根据各所述复原故障运行数据进行数据扩增，得到若干扩增故障运行数据；
    确定模块，用于获取所述目标设备对应的故障预测模型，根据若干所述扩增故障运行数据确定所述故障预测模型对应的第一训练数据集；
    训练模块，用于根据所述第一训练数据集对所述故障预测模型进行训练，根据已训练的所述故障预测模型确定所述目标设备对应的目标故障预测模型。
  9. 一种终端，其特征在于，所述终端包括有存储器和一个或者一个以上处理器；所述存储器存储有一个或者一个以上的程序；所述程序包含用于执行如权利要求1-7中任一所述的适用于极端状态的储能设备的故障预测模型生成方法的指令；所述处理器用于执行所述程序。
  10. 一种计算机可读存储介质，其上存储有多条指令，其特征在于，所述指令适用于由处理器加载并执行，以实现上述权利要求1-7任一所述的适用于极端状态的储能设备的故障预测模型生成方法的步骤。
PCT/CN2022/137740 2022-10-08 2022-12-08 一种适用于极端状态的储能设备的故障预测模型生成方法 WO2024073936A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211219495.8 2022-10-08
CN202211219495.8A CN115310562B (zh) 2022-10-08 2022-10-08 一种适用于极端状态的储能设备的故障预测模型生成方法

Publications (1)

Publication Number Publication Date
WO2024073936A1 true WO2024073936A1 (zh) 2024-04-11

Family

ID=83866354

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/137740 WO2024073936A1 (zh) 2022-10-08 2022-12-08 一种适用于极端状态的储能设备的故障预测模型生成方法

Country Status (2)

Country Link
CN (1) CN115310562B (zh)
WO (1) WO2024073936A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115310562B (zh) * 2022-10-08 2022-12-30 深圳先进技术研究院 一种适用于极端状态的储能设备的故障预测模型生成方法
CN116226676B (zh) * 2023-05-08 2023-07-21 中科航迈数控软件(深圳)有限公司 适用于极端环境的机床故障预测模型生成方法及相关设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110175707A (zh) * 2019-05-13 2019-08-27 东南大学 基于数据继承思想的电力系统暂态频率预测模型更新方法
US20200209841A1 (en) * 2018-12-27 2020-07-02 Utopus Insights, Inc. System and method for proactive handling of multiple faults and failure modes in an electrical network of energy assets
CN111368890A (zh) * 2020-02-26 2020-07-03 珠海格力电器股份有限公司 故障检测方法及装置、信息物理融合系统
CN115310562A (zh) * 2022-10-08 2022-11-08 深圳先进技术研究院 一种适用于极端状态的储能设备的故障预测模型生成方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110766059A (zh) * 2019-10-14 2020-02-07 四川西部能源股份有限公司郫县水电厂 一种变压器故障的预测方法、装置和设备
CN111738420B (zh) * 2020-06-24 2023-06-06 莫毓昌 一种基于多尺度抽样的机电设备状态数据补全与预测方法
CN112508243B (zh) * 2020-11-25 2022-09-09 国网浙江省电力有限公司信息通信分公司 电力信息系统多故障预测网络模型的训练方法及装置
CN112613584A (zh) * 2021-01-07 2021-04-06 国网上海市电力公司 一种故障诊断方法、装置、设备及存储介质
CN113837289B (zh) * 2021-09-26 2024-03-19 创新奇智(重庆)科技有限公司 模型训练方法、故障预测方法、装置及电子设备

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200209841A1 (en) * 2018-12-27 2020-07-02 Utopus Insights, Inc. System and method for proactive handling of multiple faults and failure modes in an electrical network of energy assets
CN110175707A (zh) * 2019-05-13 2019-08-27 东南大学 基于数据继承思想的电力系统暂态频率预测模型更新方法
CN111368890A (zh) * 2020-02-26 2020-07-03 珠海格力电器股份有限公司 故障检测方法及装置、信息物理融合系统
CN115310562A (zh) * 2022-10-08 2022-11-08 深圳先进技术研究院 一种适用于极端状态的储能设备的故障预测模型生成方法

Also Published As

Publication number Publication date
CN115310562B (zh) 2022-12-30
CN115310562A (zh) 2022-11-08

Similar Documents

Publication Publication Date Title
WO2024073936A1 (zh) 一种适用于极端状态的储能设备的故障预测模型生成方法
Driessen et al. Maintenance optimization under non-constant probabilities of imperfect inspections
WO2021164250A1 (zh) 一种湍流场更新方法、装置及其相关设备
US20060047512A1 (en) System, device, and methods for updating system-monitoring models
WO2024066055A1 (zh) 一种双向锂离子电池的寿命预测与状态估计方法
WO2024066052A1 (zh) 一种双向锂离子电池智能健康监控方法及系统
CN114154429A (zh) 一种储能系统的数字孪生体构建方法、装置及存储介质
CN113536676B (zh) 基于特征迁移学习的锂电池健康状况监测方法
WO2023241272A1 (zh) 基于图注意力网络的混凝土坝缺陷图像描述自动生成方法
CN111445105B (zh) 基于目标值分析的电厂在线性能诊断方法及系统
WO2020082972A1 (zh) 一种基于贝叶斯网络的锅炉运行数据监测方法和装置
CN116089870A (zh) 基于元学习的小样本条件下工业设备故障预测方法及装置
CN109142362A (zh) 一种基于bp神经网络的电厂用金属材料寿命评估方法
US20200311326A1 (en) Self-service artificial intelligence platform leveraging data-based and physics-based models for providing real-time controls and recommendations
WO2023226358A1 (zh) 一种储能电池包的荷电状态预测方法
CN113283113B (zh) 太阳电池阵发电电流预测模型训练方法、异常检测方法、设备及介质
CN113722860A (zh) 基于降阶模型的瞬态热力状态在线评估方法、设备及介质
CN117195451A (zh) 一种基于图论的桥梁监测数据修复方法
TWI747452B (zh) 以人工智慧進行案場異常偵測之智能監控之系統、方法及儲存媒體
CN114677547A (zh) 一种基于自保持表征扩展的类增量学习的图像分类方法
CN114021744A (zh) 设备的剩余使用寿命的确定方法、装置和电子设备
CN107145665A (zh) 一种巷道围岩应力建模与预测方法
AU2020100796A4 (en) Method for analyzing operation reliability of multi-state power generation system based on decision diagram
Cui et al. Data link fault location model based on machine learning
CN116414420B (zh) 禽类养殖控制系统的自动化升级方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22961293

Country of ref document: EP

Kind code of ref document: A1