WO2024073932A1 - 储能系统故障检测方法、装置及智能终端 - Google Patents

储能系统故障检测方法、装置及智能终端 Download PDF

Info

Publication number
WO2024073932A1
WO2024073932A1 PCT/CN2022/136761 CN2022136761W WO2024073932A1 WO 2024073932 A1 WO2024073932 A1 WO 2024073932A1 CN 2022136761 W CN2022136761 W CN 2022136761W WO 2024073932 A1 WO2024073932 A1 WO 2024073932A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
fault
statistic
detected
energy storage
Prior art date
Application number
PCT/CN2022/136761
Other languages
English (en)
French (fr)
Inventor
郭媛君
安钊
吴承科
杨之乐
刘祥飞
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 WO2024073932A1 publication Critical patent/WO2024073932A1/zh

Links

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02S—GENERATION OF ELECTRIC POWER BY CONVERSION OF INFRARED RADIATION, VISIBLE LIGHT OR ULTRAVIOLET LIGHT, e.g. USING PHOTOVOLTAIC [PV] MODULES
    • H02S50/00—Monitoring or testing of PV systems, e.g. load balancing or fault identification
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
    • G01D21/00—Measuring or testing not otherwise provided for
    • G01D21/02—Measuring two or more variables by means not covered by a single other subclass
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02S—GENERATION OF ELECTRIC POWER BY CONVERSION OF INFRARED RADIATION, VISIBLE LIGHT OR ULTRAVIOLET LIGHT, e.g. USING PHOTOVOLTAIC [PV] MODULES
    • H02S50/00—Monitoring or testing of PV systems, e.g. load balancing or fault identification
    • H02S50/10—Testing of PV devices, e.g. of PV modules or single PV cells
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02S—GENERATION OF ELECTRIC POWER BY CONVERSION OF INFRARED RADIATION, VISIBLE LIGHT OR ULTRAVIOLET LIGHT, e.g. USING PHOTOVOLTAIC [PV] MODULES
    • H02S50/00—Monitoring or testing of PV systems, e.g. load balancing or fault identification
    • H02S50/10—Testing of PV devices, e.g. of PV modules or single PV cells
    • H02S50/15—Testing of PV devices, e.g. of PV modules or single PV cells using optical means, e.g. using electroluminescence

Definitions

  • the present invention relates to the technical field of fault detection based on electrical digital data, and in particular to a method, device and intelligent terminal for detecting faults in an energy storage system.
  • the energy storage system is a part of the object or space range that is demarcated to determine the research object when analyzing the energy storage process. It includes the input and output of energy and matter, and the energy conversion and storage equipment. If the energy storage system fails, it may affect the user's electricity experience or even cause danger. Therefore, it is necessary to detect the fault of the energy storage system so that the fault can be handled in time.
  • the fault condition of the energy storage system can usually only be judged based on a single data indicator. For example, the measured voltage is compared with a preset voltage threshold to determine whether the energy storage system has a fault.
  • the problem with the prior art is that the fault of the energy storage system can only be judged based on a single data indicator, and it is impossible to combine multiple information for fault detection, which is not conducive to improving the accuracy of fault detection.
  • the main purpose of the present invention is to provide a method, device and intelligent terminal for detecting faults in an energy storage system, aiming to solve the problem that the prior art can only judge the faults of the energy storage system based on a single data indicator, and cannot combine multiple information for fault detection, which is not conducive to improving the accuracy of fault detection.
  • the first aspect of the present invention provides a method for detecting a fault in an energy storage system, wherein the method for detecting a fault in an energy storage system comprises:
  • Determining whether the data to be detected is fault data according to a first size relationship and/or a second size relationship, wherein the first size relationship is a size relationship between the T2 statistic limit and the T2 statistic, and the second size relationship is a size relationship between the Q statistic limit and the Q statistic;
  • the trained fault detection model is used to perform fault detection on the data to be detected and obtain the fault category;
  • the above T2 statistic is the Hotelling statistic
  • the above Q statistic is the squared prediction error statistic
  • the above T2 statistic limit is the Hotelling statistic limit
  • the above Q statistic limit is the squared prediction error statistic limit.
  • the above-mentioned power information change data includes voltage data, current data, temperature data and energy storage change data
  • the above-mentioned equipment component image data includes surface image data of solar cells in the above-mentioned energy storage system and surface image data of wind equipment blades.
  • the method before acquiring at least one set of fault-free data corresponding to the energy storage system based on the pre-built digital twin system, the method further includes:
  • each group of raw data includes power information change data, equipment component image data, and fault labeling information of the energy storage system within a collection period, and the fault labeling information is used to indicate whether the raw data is fault data or non-fault data;
  • the above original data and the above amplified data are uploaded to the digital twin database, and a digital twin system corresponding to the above energy storage system is constructed.
  • the statistical features include a mean and a standard deviation, and the T2 statistic and Q statistic corresponding to the data to be detected are obtained by calculating according to the data to be detected and the statistical features, including:
  • the T2 statistic and the Q statistic corresponding to the above-mentioned data to be detected are calculated according to the above-mentioned standard data to be detected.
  • performing fault detection on the data to be detected by using a trained fault detection model and obtaining a fault category includes:
  • the above-mentioned data to be detected is input into the above-mentioned trained fault detection model, and the fault category corresponding to the above-mentioned data to be detected is output through the above-mentioned trained fault detection model, wherein the above-mentioned trained fault detection model performs fault classification on the power information change data in the above-mentioned data to be detected based on the logistic regression method, and the above-mentioned trained fault detection model performs fault classification on the equipment component image data in the above-mentioned data to be detected based on VGG-Net.
  • the above-trained fault detection model is trained according to the following steps:
  • the above-mentioned training data includes multiple groups of training information groups, each group of the above-mentioned training information group includes power information training data, equipment component image training data and marked fault categories;
  • the model parameters of the above-mentioned fault detection model are adjusted according to the above-mentioned detected fault category and the corresponding above-mentioned labeled fault category, and the above-mentioned step of inputting the power information training data and the equipment component image training data in the training data into the above-mentioned fault detection model is continued until the preset training conditions are met to obtain a trained fault detection model.
  • the method further includes:
  • faults are marked in the above digital twin system and the marking results are visualized and output.
  • the method further includes:
  • a second aspect of the present invention provides an energy storage system fault detection device, wherein the energy storage system fault detection device comprises:
  • a fault-free data acquisition module used to acquire at least one set of fault-free data corresponding to the energy storage system based on a pre-built digital twin system, wherein the fault-free data includes power information change data and equipment component image data of the energy storage system during a fault-free period;
  • a statistical limit value calculation module is used to perform principal component analysis on the above-mentioned fault-free data to calculate and obtain the T2 statistical limit value, Q statistical limit value and statistical characteristics corresponding to the above-mentioned fault-free data;
  • a data processing module for detection is used to obtain a set of data to be detected corresponding to the energy storage system, and calculate the T2 statistic and Q statistic corresponding to the data to be detected according to the data to be detected and the statistical characteristics, wherein the data to be detected includes the power information change data and equipment component image data of the energy storage system within the detection period;
  • a fault data judgment module used for judging whether the data to be detected is fault data according to a first size relationship and/or a second size relationship, wherein the first size relationship is the size relationship between the T2 statistic limit and the T2 statistic, and the second size relationship is the size relationship between the Q statistic limit and the Q statistic;
  • a fault category detection module used for, when the data to be detected is fault data, performing fault detection on the data to be detected by using a trained fault detection model and obtaining a fault category;
  • the above T2 statistic is the Hotelling statistic
  • the above Q statistic is the squared prediction error statistic
  • the above T2 statistic limit is the Hotelling statistic limit
  • the above Q statistic limit is the squared prediction error statistic limit.
  • a third aspect of the present invention provides an intelligent terminal, which includes a memory, a processor, and an energy storage system fault detection program stored in the memory and executable on the processor.
  • the energy storage system fault detection program is executed by the processor, the steps of any one of the above-mentioned energy storage system fault detection methods are implemented.
  • At least one set of fault-free data corresponding to the energy storage system is obtained based on the pre-built digital twin system, wherein the fault-free data includes the power information change data and equipment component image data of the energy storage system in the fault-free period; the fault-free data is subjected to principal component analysis, and the T2 statistic limit, Q statistic limit and statistical characteristics corresponding to the fault-free data are calculated; a set of data to be detected corresponding to the energy storage system is obtained, and the T2 statistic and Q statistic corresponding to the data to be detected are calculated according to the data to be detected and the statistical characteristics, wherein the data to be detected includes the power information change data and equipment component image data of the energy storage system in the period to be detected; whether the data to be detected is fault data is determined according to the first size relationship and/or the second size relationship, wherein the first size relationship is the size relationship between the T2 statistic limit and the T2 statistic, and the second size relationship is the size relationship between the Q statistic limit and the Q
  • a set of data to be detected in the present invention is composed of multiple data, specifically including the power information change data and equipment component image data of the energy storage system in the time period to be detected.
  • the T2 statistic limit and the Q statistic limit are first calculated based on a set of fault-free data corresponding to the energy storage system, and the T2 statistic and Q statistic are calculated for the data to be detected. According to the size relationship between the corresponding statistics and the statistic limit, it can be analyzed and judged whether the data to be detected is fault data. When it is fault data, the above data to be detected is further detected by the trained fault detection model and the fault category is obtained.
  • multiple information in the data to be detected (including power information change data and equipment component image data) can be combined for fault judgment and detection, which is conducive to improving the accuracy of fault detection.
  • FIG1 is a schematic flow chart of a method for detecting a fault in an energy storage system according to an embodiment of the present invention
  • FIG2 is a schematic diagram of a process for building a digital twin system provided by an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of the structure of an energy storage system fault detection device provided by an embodiment of the present invention.
  • FIG. 4 is a block diagram of the internal structure of a smart terminal provided by an embodiment of the present invention.
  • the term “if” may be interpreted as “when” or “upon” or “in response to determining” or “in response to detecting,” depending on the context.
  • the phrases “if it is determined” or “if [described condition or event] is detected” may be interpreted as meaning “upon determination” or “in response to determining” or “upon detection of [described condition or event]” or “in response to detecting [described condition or event],” depending on the context.
  • the energy storage system is a part of the object or space range that is demarcated to determine the research object when analyzing the energy storage process. It includes the input and output of energy and matter, and the energy conversion and storage equipment. If the energy storage system fails, it may affect the user's electricity experience or even cause danger. Therefore, it is necessary to detect the fault of the energy storage system so that the fault can be handled in time.
  • the fault condition of the energy storage system can usually only be judged based on a single data indicator. For example, the measured voltage is compared with a preset voltage threshold to determine whether the energy storage system has a fault.
  • the problem with the prior art is that the fault of the energy storage system can only be judged based on a single data indicator, and it is impossible to combine multiple information for fault detection, which is not conducive to improving the accuracy of fault detection.
  • the existing technology can only process one form of fault data at a time, and the original training data (or fault data used for training) is small, making it difficult to conduct effective training.
  • the efficiency of human-computer interaction is low, and workers are often dispatched to handle the problem only after the failure occurs, and the workers must be experienced and know which part has failed based on experience.
  • At least one group of fault-free data corresponding to the energy storage system is obtained based on a pre-built digital twin system, wherein the fault-free data includes the power information change data and equipment component image data of the energy storage system within the fault-free period; the fault-free data is subjected to principal component analysis to calculate and obtain the T2 statistic limit, Q statistic limit and statistical characteristics corresponding to the fault-free data; a group of data to be detected corresponding to the energy storage system is obtained, and the T2 statistic and Q statistic corresponding to the data to be detected are calculated according to the data to be detected and the statistical characteristics, wherein the data to be detected includes the power information change data and equipment component image data of the energy storage system within the period to be detected; and whether the data to be detected is fault data is determined according to the first size relationship and/or the second size relationship, wherein the first size relationship is the relationship between the T2 statistic limit and the T The size relationship between the T 2 statistics is
  • a set of data to be detected in the present invention is composed of multiple data, specifically including the power information change data and equipment component image data of the energy storage system in the time period to be detected.
  • the T2 statistic limit and the Q statistic limit are first calculated based on a set of fault-free data corresponding to the energy storage system, and the T2 statistic and Q statistic are calculated for the data to be detected. According to the size relationship between the corresponding statistics and the statistic limit, it can be analyzed and judged whether the data to be detected is fault data. When it is fault data, the above data to be detected is further detected by the trained fault detection model and the fault category is obtained.
  • multiple information in the data to be detected (including power information change data and equipment component image data) can be combined for fault judgment and detection, which is conducive to improving the accuracy of fault detection.
  • fault detection is not performed on all data to be detected through the fault detection model, but the data with faults is first determined and then further fault detection is performed through the fault detection model, which can reduce the amount of calculation and improve the efficiency of fault detection.
  • the training data of the fault detection model can also be only the data with faults, which can improve the training speed of the fault detection model.
  • the present invention can also perform sample amplification on the collected raw data, thereby increasing the amount of data available for training.
  • tracking and positioning can be achieved based on the digital twin system, and problems can be quickly solved.
  • Fault visualization and simulation adjustments can be made for faults, thereby improving the efficiency of fault resolution when faults occur.
  • an embodiment of the present invention provides a method for detecting a fault in an energy storage system. Specifically, the method comprises the following steps:
  • Step S100 obtaining at least one set of fault-free data corresponding to the energy storage system based on a pre-built digital twin system, wherein the fault-free data includes power information change data and equipment component image data of the energy storage system during a fault-free period.
  • the digital twin system is constructed in advance based on the energy storage system for monitoring and fault detection as needed.
  • the above-mentioned fault-free data is a set of data corresponding to the energy storage system when there is no fault. It should be noted that the energy storage system does not have a fault means that the energy storage system can operate normally according to pre-defined data indicators (such as defined voltage, current, power, etc.), and there is no damage to the equipment in the energy storage system.
  • the above-mentioned fault-free period is any time period of a preset time length in which there is no fault; in another application scenario, the above-mentioned fault-free period is a time period of a preset time length in which there is no fault, which is pre-set or selected by the user, and is not specifically limited here. It should be noted that the time length of the above-mentioned fault-free period is pre-set, and the time length of the fault-free period is the same as the time length of the period to be detected.
  • the above-mentioned power information change data is a time series signal within a period of time (such as a fault-free period), that is, it is obtained by continuous multiple data collection based on the digital twin system within this period
  • the above-mentioned equipment component image data is a two-dimensional visual signal (which can be a depth image or a color image) obtained by collecting images of equipment components in the energy storage system at a moment in this period (in this embodiment, the current moment of fault detection).
  • the above-mentioned equipment component image data can be obtained based on equipment components in the digital twin system, or it can be obtained by collecting images of equipment components in the actual energy storage system, which is not specifically limited here.
  • the fault-free data and the data to be detected in this embodiment are both composed of power information change data and equipment component image data, that is, they are composed of heterogeneous signals obtained by two different means, and then fault detection can be performed based on multi-source heterogeneous data and a variety of information, thereby improving the accuracy of fault detection.
  • the above-mentioned power information change data includes voltage data, current data, temperature data and energy storage change data
  • the above-mentioned equipment component image data includes the surface image data of solar cells and the surface image data of wind power equipment blades in the above-mentioned energy storage system.
  • the above-mentioned energy storage system is a new energy storage system based on wind power generation and solar power generation, so the main focus is on the surface image data of solar cells and wind power equipment, so as to facilitate the judgment of whether there are faults such as solar cell panel rupture and wind turbine blade cracks.
  • the above-mentioned power information change data may also include voltage difference data, energy storage density data, energy storage power data, energy storage efficiency data, etc., which are not specifically limited here.
  • FIG2 is a schematic diagram of a process for building a digital twin system provided by an embodiment of the present invention. As shown in FIG2 , in this embodiment, before the above step S100, the above energy storage system fault detection method further includes:
  • Step A100 collecting and acquiring multiple groups of original data corresponding to the above energy storage system, wherein each group of the above original data includes power information change data, equipment component image data and fault marking information of the above energy storage system within a collection period, and the above fault marking information is used to indicate whether the above original data is fault data or non-fault data.
  • Step A200 performing sample amplification based on the original data to obtain multiple groups of amplified data, wherein different amplification channels are used to perform sample amplification on the power information change data and the device component image data.
  • Step A300 upload the above-mentioned original data and the above-mentioned amplified data to the digital twin database, and construct a digital twin system corresponding to the above-mentioned energy storage system.
  • Digital twins are simulation processes that make full use of data such as physical models, sensor updates, and operation history, and integrate multiple disciplines, multiple physical quantities, multiple scales, and multiple probabilities.
  • the digital twin system in this embodiment is a digital mapping system corresponding to the above energy storage system.
  • the digital twin system is constructed based on the devices in the energy storage system, the relationships between the devices, the device parameters, the operating status of the devices, etc.
  • the digital twin system can also interact with the energy storage system for data.
  • each group of raw data may be data with a fault or data without a fault, and is distinguished and determined according to the fault labeling information in the raw data.
  • Each group of raw data is collected and obtained in a different collection period, but the specific time length of the collection period is the same, and the specific time length of the collection period is the same as the specific time length of the above-mentioned fault-free period.
  • sample amplification is also performed based on the original data to increase the amount of data in the digital twin system.
  • sample amplification can be performed for each set of original data to obtain one or more sets of corresponding amplified data.
  • the power information change data is time series data
  • the equipment component image data is image data. The two have different characteristics. Therefore, in this embodiment, different methods are used to amplify the two types of data respectively to obtain better data amplification results.
  • a dual-channel mode is used to amplify the original data to solve the problem of insufficient data volume, especially the problem of random fault occurrence.
  • the generative adversarial network GAN
  • DCGAN deep convolution generative adversarial network
  • more fault data can also be obtained based on sample amplification, which is conducive to providing more training data for the training process of the fault detection model.
  • the amplified data includes the amplified power information change data, equipment component image data, and corresponding fault labeling information.
  • the amplified data is obtained by fitting in different ways based on the dual channels (for example, smoothing and fitting for images, and curve fitting for time series signals).
  • the fault labeling information of the amplified data is the same as the corresponding original data.
  • the original data is processed in different ways based on the dual channels, and fault noise is added, so that amplified fault data can be obtained.
  • All the generated amplified data and original data are uploaded to the digital twin database to generate an energy storage system in the virtual world that is the same as in the real world.
  • Step S200 performing principal component analysis on the above-mentioned fault-free data, and calculating and obtaining T2 statistic limit value, Q statistic limit value and statistical characteristics corresponding to the above-mentioned fault-free data.
  • dual-channel principal component analysis is performed on the above-mentioned fault-free data, that is, the power information change data and the equipment component image data are respectively subjected to principal component analysis, and the T2 statistic limit, Q statistic limit and statistical characteristics corresponding to the power information change data and the equipment component image data are respectively obtained.
  • the equipment component image data can also be processed into numerical data in the same form as the power information change data, so as to be subjected to principal component analysis together with the power information change data, which is not specifically limited here.
  • the corresponding statistical limit values can be calculated based on one or more groups of acquired fault-free data.
  • the calculation of one group of fault-free data is taken as an example for explanation.
  • the corresponding T2 statistical limit values, Q statistical limit values and statistical characteristics can be calculated for each group of fault-free data respectively, and the subsequent steps of calculating the T2 statistic and the Q statistic are respectively executed according to the calculated two groups of statistical characteristics.
  • the size relationship of multiple groups is comprehensively considered to determine whether the data to be detected is fault data.
  • the principal component analysis of a data is used as an example for explanation, but it is not a specific limitation.
  • the dimensionality reduction is performed based on the principal component analysis (PCA) technology.
  • PCA principal component analysis
  • the relationship between the statistical limit and the corresponding statistical quantity can be used to determine whether the corresponding data has a fault.
  • the specific statistical features are not limited.
  • the statistical features may include a mean and a standard deviation. In this case, when the data to be detected is processed, each feature data in the data to be detected is standardized based on the mean and the standard deviation.
  • the statistical features may include a mean and a root mean square. In this case, when the data to be detected is processed, each feature data in the data to be detected is standardized based on the mean and the root mean square.
  • the statistical features including the mean and the root mean square are taken as an example for explanation, specifically including the mean and standard deviation of each feature in the above-mentioned fault-free data (ie, each type of power information change data and each type of equipment component image data).
  • fault-free data may be preprocessed and normalized first, and the data with a normalized mean of 0 and a root mean square of 1 for each feature thereof may be normalized to reduce the amount of calculation and improve processing efficiency.
  • the fault-free data is a matrix X * n * m with n rows and m columns, where each element is represented by X * ij , i is an integer from 1 to n, j is an integer from 1 to m, n represents the number of samples, and m represents the sample dimension (i.e., the number of features).
  • each feature in the fault-free data is normalized to data with a mean of 0 and a root mean square of 1 to obtain normalized fault-free data.
  • a preset ratio threshold e.g. 70%
  • the T2 statistic limit and the Q statistic limit are calculated respectively according to the number of selected features and the pre-set statistic limit calculation formula.
  • the specific statistic limit calculation formula and calculation process can refer to the solution formula and solution process of the corresponding statistic limit in the PCA dimensionality reduction process in the prior art, and are not specifically limited here.
  • Step S300 obtaining a set of data to be detected corresponding to the energy storage system, and calculating T2 statistics and Q statistics corresponding to the data to be detected based on the data to be detected and the statistical characteristics, wherein the data to be detected includes power information change data and equipment component image data of the energy storage system during the period to be detected.
  • the above-mentioned data to be detected is a set of data obtained by collecting data from the energy storage system during the operation of the energy storage system.
  • the above-mentioned data to be detected can also be system operation prediction data for a period of time obtained after a certain operation is performed on the digital twin system (to predict whether this operation will cause a fault), which is not specifically limited here.
  • the time length of the data to be detected is the same as the time length of the fault-free data, and the data format is also the same, and the time length of the detection period is the same as the time length of the fault-free period.
  • the above-mentioned method of calculating the T2 statistic and Q statistic corresponding to the above-mentioned data to be detected based on the above-mentioned data to be detected and the above-mentioned statistical characteristics includes: standardizing the above-mentioned data to be detected according to the above-mentioned statistical characteristics to obtain standard data to be detected; and calculating the T2 statistic and Q statistic corresponding to the above-mentioned data to be detected according to the above-mentioned standard data to be detected.
  • the above-mentioned fault-free data is first standardized, so the corresponding statistical features have a mean value of 0 and a root mean square value of 1. Based on this, the data to be detected is standardized, that is, the mean and root mean square of the data to be detected are also processed to 0 and 1 to obtain the corresponding standard data to be detected.
  • the above T 2 statistic and Q statistic can be calculated according to the preset statistic calculation formula.
  • the specific statistic calculation formula and calculation process can refer to the solution formula and solution process of the corresponding statistic in the PCA dimensionality reduction process in the prior art, and are not specifically limited here. It should be noted that in the calculation process of the T 2 statistic and the Q statistic, the number of principal components used is the same as the number of principal components k used in the calculation of the statistic limit above.
  • Step S400 determining whether the data to be detected is fault data according to the first size relationship and/or the second size relationship.
  • the above-mentioned first size relationship is the size relationship between the above-mentioned T2 statistic limit and the above-mentioned T2 statistic
  • the above-mentioned second size relationship is the size relationship between the above-mentioned Q statistic limit and the above-mentioned Q statistic.
  • the data to be detected is fault data; otherwise, the data to be detected is not fault data.
  • the data to be detected is fault data, otherwise the data to be detected is not fault data.
  • the first statistic supplement value and the second statistic supplement value can be preset and adjusted according to actual needs.
  • PCA is first used for fault detection, and normal data is used to train PCA to obtain the T2 statistic limit and Q statistic limit.
  • the data to be detected is used for testing, the T2 statistic and Q statistic of the data to be detected are obtained. If one of the statistics exceeds the corresponding statistic threshold, it is considered that the data to be detected has a fault, and the fault category will be further detected.
  • Step S500 When the data to be detected is fault data, a fault detection is performed on the data to be detected by using a trained fault detection model to obtain a fault category.
  • the above T 2 statistic is the Hotelling statistic
  • the above Q statistic is the squared prediction error (SPE) statistic
  • the above T 2 statistic limit is the Hotelling statistic limit
  • the above Q statistic limit is the squared prediction error statistic limit.
  • the Hotelling statistic, the squared prediction error statistic, the Hotelling statistic limit, and the squared prediction error statistic limit are statistics or limits calculated in the process of principal component analysis.
  • the specific calculation formula and calculation process can refer to the solution formula and solution process of the corresponding statistic in the PCA dimensionality reduction process in the prior art, and are not specifically limited here.
  • the above-mentioned trained fault detection model is a pre-trained fault detection model.
  • the above-mentioned data to be detected includes two different types of data, so the above-mentioned fault detection model can use two different channels to process them respectively, thereby realizing dual-channel fault classification.
  • the above step S500 specifically includes: inputting the above data to be detected into the above trained fault detection model, and outputting the fault category corresponding to the above data to be detected through the above trained fault detection model, wherein the above trained fault detection model classifies the power information change data in the above data to be detected based on the logistic regression method, and the above trained fault detection model classifies the equipment component image data in the above data to be detected based on VGG-Net.
  • the above two parts including the logistic regression part and the VGG-Net part
  • the above fault detection model may include two sub-models, which process power information change data and device component image data respectively.
  • two separately trained sub-models may also be used to form the above fault detection model, but this is not a specific limitation.
  • VGG-net is used to classify the faults and determine which type of fault it is.
  • the above fault categories can be pre-set according to actual needs.
  • the fault categories may include voltage instability, excessive current, solar panel breakage, wind blade breakage, damage to other specific components or equipment, etc., which are not specifically limited here.
  • the above-mentioned trained fault detection model is trained according to the following steps:
  • the above-mentioned training data includes multiple groups of training information groups, each group of the above-mentioned training information group includes power information training data, equipment component image training data and marked fault categories;
  • the model parameters of the above-mentioned fault detection model are adjusted according to the above-mentioned detected fault category and the corresponding above-mentioned labeled fault category, and the above-mentioned step of inputting the power information training data and the equipment component image training data in the training data into the above-mentioned fault detection model is continued until the preset training conditions are met to obtain a trained fault detection model.
  • the fault detection model only needs to perform category detection on data with faults, so the training data may only include data with faults, thereby reducing the amount of calculation.
  • the above-mentioned fault detection model can process two types of data (power information training data and equipment component image training data) respectively, so the two types of data can also have corresponding fault categories respectively.
  • the above-mentioned detection fault category can include a first detection fault category and a second detection fault category
  • the above-mentioned labeling fault category can include a first labeling fault category and a second labeling fault category.
  • the first detection fault category corresponds to the first labeling fault category
  • the second detection fault category corresponds to the second labeling fault category.
  • the corresponding loss values can be calculated according to the first detection fault category and the first labeling fault category, and the second detection fault category and the second labeling fault category, and the model parameters are adjusted according to the loss values. It should be noted that different thresholds can also be set for the losses corresponding to the two different data to flexibly adjust the degree of attention of the two data.
  • the above-mentioned preset training condition is that the number of iterations during training reaches a preset iteration threshold, or the calculated loss value is less than a preset loss threshold.
  • the above-mentioned method further includes: marking the fault in the above-mentioned digital twin system based on the above-mentioned fault category and visually outputting the marking result.
  • the digital twin system can simultaneously process two different forms of fault data, namely time series and pictures, in the energy storage system. And the data is subjected to fault detection, classified and uploaded to the digital twin database in real time, and the real-time display of whether there is a fault and what kind of fault it is in the virtual world.
  • the relationship between the preset fault category and the damaged parts for example, a pre-established fault category and damaged parts relationship table
  • the above-mentioned method further includes: obtaining a fault adjustment strategy based on the above-mentioned fault category; inputting the above-mentioned fault adjustment strategy into the above-mentioned digital twin system, and executing the above-mentioned fault adjustment strategy in the above-mentioned digital twin system; obtaining and outputting the fault adjustment result in the above-mentioned digital twin system.
  • the above-mentioned fault adjustment strategy can be input by the target object (i.e., the operator), or can be searched and obtained in a pre-set adjustment strategy table according to the above-mentioned fault category, or can be searched and obtained in the cloud, which is not specifically limited here.
  • the above-mentioned fault adjustment strategy is a specific scheme for adjusting the equipment in the energy storage system according to the fault category. After obtaining the fault adjustment strategy, it is input into the digital twin system. In this way, the strategy can be simulated and executed in the digital twin system first to obtain the fault adjustment result.
  • the corresponding strategy can also be executed in the actual energy storage system. Otherwise, a new fault adjustment strategy is re-obtained and re-executed in the digital twin system. In this way, the efficiency of fault adjustment can be effectively improved based on the digital twin system, the processing difficulty can be reduced, and misoperation can be avoided.
  • a group of data to be detected is composed of a combination of multiple data, specifically including power information change data and equipment component image data of the energy storage system in the time period to be detected.
  • the T2 statistic limit and Q statistic limit are first calculated based on a group of fault-free data corresponding to the energy storage system, and the T2 statistic and Q statistic are calculated for the data to be detected. According to the size relationship between the corresponding statistics and the statistic limit, it can be analyzed and judged whether the data to be detected is fault data. When it is fault data, the above data to be detected is further detected by the trained fault detection model and the fault category is obtained.
  • multiple information in the data to be detected (including power information change data and equipment component image data) can be combined for fault judgment and detection, which is conducive to improving the accuracy of fault detection.
  • fault detection is not performed on all data to be detected through the fault detection model, but the data with faults is first determined and then further fault detection is performed through the fault detection model, which can reduce the amount of calculation and improve the efficiency of fault detection.
  • the training data of the fault detection model can also be only the data with faults, which can improve the training speed of the fault detection model.
  • the collected raw data can also be sampled to increase the amount of data available for training.
  • a digital twin system can be built, based on which tracking and positioning can be achieved, problems can be quickly solved, faults can be visualized, and simulated and adjusted for faults, thereby improving the efficiency of fault resolution when a fault occurs.
  • an embodiment of the present invention further provides an energy storage system fault detection device, and the above energy storage system fault detection device includes:
  • the fault-free data acquisition module 610 is used to acquire at least one set of fault-free data corresponding to the energy storage system based on the pre-built digital twin system, wherein the fault-free data includes power information change data and equipment component image data of the energy storage system during the fault-free period.
  • the statistic limit value calculation module 620 is used to perform principal component analysis on the above-mentioned fault-free data, and calculate and obtain the T2 statistic limit value, Q statistic limit value and statistical characteristics corresponding to the above-mentioned fault-free data.
  • the data processing module 630 is used to obtain a set of data to be detected corresponding to the above energy storage system, and calculate the T2 statistic and Q statistic corresponding to the above data to be detected based on the above data to be detected and the above statistical characteristics, wherein the above data to be detected includes the power information change data and equipment component image data of the above energy storage system during the period to be detected.
  • the fault data judgment module 640 is used to judge whether the above-mentioned data to be detected is fault data according to the first size relationship and/or the second size relationship, wherein the above-mentioned first size relationship is the size relationship between the above-mentioned T2 statistic limit and the above-mentioned T2 statistic, and the above-mentioned second size relationship is the size relationship between the above-mentioned Q statistic limit and the above-mentioned Q statistic.
  • the fault category detection module 650 is used to perform fault detection on the data to be detected by using a trained fault detection model and obtain a fault category when the data to be detected is fault data.
  • the above statistic is the Hotelling statistic
  • the above Q statistic is the squared prediction error (SPE) statistic
  • the above T2 statistic limit is the Hotelling statistic limit
  • the above Q statistic limit is the squared prediction error statistic limit.
  • the Hotelling statistic, the squared prediction error statistic, the Hotelling statistic limit, and the squared prediction error statistic limit are statistics or limits calculated in the process of principal component analysis.
  • the specific calculation formula and calculation process can refer to the solution formula and solution process of the corresponding statistic in the PCA dimensionality reduction process in the prior art, and are not specifically limited here.
  • the specific functions of the above energy storage system fault detection device and its modules can refer to the corresponding description in the above energy storage system fault detection method, which will not be repeated here.
  • the present invention also provides an intelligent terminal, whose principle block diagram can be shown in Figure 4.
  • the above intelligent terminal includes a processor and a memory.
  • the memory of the intelligent terminal includes an energy storage system fault detection program, and the memory provides an environment for the operation of the energy storage system fault detection program.
  • the energy storage system fault detection program is executed by the processor, the steps of any of the above energy storage system fault detection methods are implemented.
  • the above intelligent terminal may also include other functional modules or units, which are not specifically limited here.
  • FIG. 4 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 smart terminal to which the solution of the present invention is applied.
  • the smart terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
  • An embodiment of the present invention further provides a computer-readable storage medium, on which an energy storage system fault detection program is stored.
  • the energy storage system fault detection program is executed by a processor, the steps of any one of the energy storage system fault detection methods provided in the embodiments of the present invention are implemented.
  • the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the above-mentioned device can be divided into different functional units or modules to complete all or part of the functions described above.
  • the functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit.
  • the above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units.
  • the disclosed device/intelligent terminal and method can be implemented in other ways.
  • the device/intelligent terminal embodiments described above are only schematic, for example, the division of the above modules or units is only a logical function division, and in actual implementation, other division methods can be used, for example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.
  • the above integrated modules/units are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above embodiment method, and can also be completed by instructing the relevant hardware through a computer program.
  • the above computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above method embodiments can be implemented.
  • the above computer program includes computer program code, and the above computer program code can be in source code form, object code form, executable file or some intermediate form.
  • the above computer-readable medium can include: any entity or device capable of carrying the above computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the above computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Supply And Distribution Of Alternating Current (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

本发明公开了一种储能系统故障检测方法、装置及智能终端，方法包括：基于预先构建的数字孪生系统获取储能系统的至少一组无故障数据；对无故障数据进行主成分分析处理获得无故障数据对应的T 2统计量限值、Q统计量限值及统计特征；获取储能系统的一组待检测数据，根据待检测数据和统计特征计算获得待检测数据对应的T 2统计量和Q统计量，待检测数据包括储能系统在待检测时段内的电力信息变化数据和设备部件图像数据；根据第一大小关系和/或第二大小关系判断待检测数据是否为故障数据；待检测数据是故障数据时，通过已训练的故障检测模型对待检测数据进行故障检测并获得故障类别。本发明方案有利于提高故障检测的准确性。

Description

储能系统故障检测方法、装置及智能终端 技术领域
本发明涉及基于电数字数据的故障检测技术领域，尤其涉及的是一种储能系统故障检测方法、装置及智能终端。
 背景技术
随着科学技术的发展，对于储能系统的研究和控制越来越受到重视。储能系统是在对储能过程进行分析时，为了确定研究对象而划出的部分物体或空间范围。它包括能量和物质的输入和输出、能量的转换和储存设备。如果储能系统出现故障，则有可能影响用户的用电体验甚至引发危险，因此需要对储能系统的故障情况进行检测，以便及时针对故障进行处理。
储能系统往往涉及多种能量、多种设备等，是随时间变换的复杂系统。而现有技术中，通常只能根据单一的数据指标对储能系统的故障情况进行判断，例如，将测量获得的电压与预设的电压阈值进行比较从而判断储能系统是否发生故障。现有技术的问题在于，只能基于单一的数据指标对储能系统的故障进行判断，无法结合多种信息进行故障检测，不利于提高故障检测的准确性。
因此，现有技术还有待改进和发展。
 技术问题
本发明的主要目的在于提供一种储能系统故障检测方法、装置及智能终端，旨在解决现有技术中只能基于单一的数据指标对储能系统的故障进行判断，无法结合多种信息进行故障检测，不利于提高故障检测的准确性的问题。
 技术解决方案
为了实现上述目的，本发明第一方面提供一种储能系统故障检测方法，其中，上述储能系统故障检测方法包括：
基于预先构建的数字孪生系统获取储能系统对应的至少一组无故障数据，其中，上述无故障数据包括上述储能系统在无故障时段内的电力信息变化数据和设备部件图像数据；
对上述无故障数据进行主成分分析处理，计算获得上述无故障数据对应的T 2统计量限值、Q统计量限值以及统计特征；
获取上述储能系统对应的一组待检测数据，根据上述待检测数据和上述统计特征计算获得上述待检测数据对应的T 2统计量和Q统计量，其中，上述待检测数据包括上述储能系统在待检测时段内的电力信息变化数据和设备部件图像数据；
根据第一大小关系和/或第二大小关系判断上述待检测数据是否为故障数据，其中，上述第一大小关系是上述T 2统计量限值与上述T 2统计量之间的大小关系，上述第二大小关系是上述Q统计量限值与上述Q统计量之间的大小关系；
当上述待检测数据是故障数据时，通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别；
其中，上述T 2统计量是霍特林统计量，上述Q统计量是平方预测误差统计量，上述T 2统计量限值是霍特林统计量限值，上述Q统计量限值是平方预测误差统计量限值。
可选的，上述电力信息变化数据包括电压数据、电流数据、温度数据和储能变化数据，上述设备部件图像数据包括上述储能系统中太阳能电池表面图像数据和风力设备叶片表面图像数据。
可选的，在上述基于预先构建的数字孪生系统获取储能系统对应的至少一组无故障数据之前，上述方法还包括：
采集获取上述储能系统对应的多组原始数据，其中，每一组上述原始数据包括上述储能系统在一个采集时段内的电力信息变化数据、设备部件图像数据以及故障标注信息，上述故障标注信息用于指示上述原始数据是故障数据或无故障数据；
基于上述原始数据进行样本扩增获得多组扩增数据，其中，对上述电力信息变化数据和上述设备部件图像数据采用不同的扩增通道进行样本扩增；
将上述原始数据和上述扩增数据上传到数字孪生数据库，并构建与上述储能系统对应的数字孪生系统。
可选的，上述统计特征包括均值和标准差，上述根据上述待检测数据和上述统计特征计算获得上述待检测数据对应的T 2统计量和Q统计量，包括：
根据上述统计特征对上述待检测数据进行标准化处理，获得待检测标准数据；
根据上述待检测标准数据计算获取上述待检测数据对应的T 2统计量和Q统计量。
可选的，上述当上述待检测数据是故障数据时，通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别，包括：
将上述待检测数据输入上述已训练的故障检测模型，通过上述已训练的故障检测模型输出上述待检测数据对应的故障类别，其中，上述已训练的故障检测模型基于逻辑回归方法对上述待检测数据中的电力信息变化数据进行故障分类，上述已训练的故障检测模型基于VGG-Net对上述待检测数据中的设备部件图像数据进行故障分类。
可选的，上述已训练的故障检测模型根据如下步骤进行训练：
将训练数据中的电力信息训练数据和设备部件图像训练数据输入上述故障检测模型，通过上述故障检测模型对上述电力信息训练数据和上述设备部件图像训练数据进行故障分类并获得检测故障类别，其中，上述训练数据包括多组训练信息组，每一组上述训练信息组包括电力信息训练数据、设备部件图像训练数据以及标注故障类别；
根据上述检测故障类别和对应的上述标注故障类别对上述故障检测模型的模型参数进行调整，并继续执行上述将训练数据中的电力信息训练数据和设备部件图像训练数据输入上述故障检测模型的步骤，直至满足预设训练条件，以得到已训练的故障检测模型。
可选的，在上述通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别之后，上述方法还包括：
基于上述故障类别在上述数字孪生系统中进行故障标记并将标记结果进行可视化输出。
可选的，在上述通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别之后，上述方法还包括：
基于上述故障类别获取故障调整策略；
将上述故障调整策略输入上述数字孪生系统，并在上述数字孪生系统中执行上述故障调整策略；
获取上述数字孪生系统中的故障调整结果并输出。
本发明第二方面提供一种储能系统故障检测装置，其中，上述储能系统故障检测装置包括：
无故障数据获取模块，用于基于预先构建的数字孪生系统获取储能系统对应的至少一组无故障数据，其中，上述无故障数据包括上述储能系统在无故障时段内的电力信息变化数据和设备部件图像数据；
统计量限值计算模块，用于对上述无故障数据进行主成分分析处理，计算获得上述无故障数据对应的T 2统计量限值、Q统计量限值以及统计特征；
待检测数据处理模块，用于获取上述储能系统对应的一组待检测数据，根据上述待检测数据和上述统计特征计算获得上述待检测数据对应的T 2统计量和Q统计量，其中，上述待检测数据包括上述储能系统在待检测时段内的电力信息变化数据和设备部件图像数据；
故障数据判断模块，用于根据第一大小关系和/或第二大小关系判断上述待检测数据是否为故障数据，其中，上述第一大小关系是上述T 2统计量限值与上述T 2统计量之间的大小关系，上述第二大小关系是上述Q统计量限值与上述Q统计量之间的大小关系；
故障类别检测模块，用于当上述待检测数据是故障数据时，通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别；
其中，上述T 2统计量是霍特林统计量，上述Q统计量是平方预测误差统计量，上述T 2统计量限值是霍特林统计量限值，上述Q统计量限值是平方预测误差统计量限值。
本发明第三方面提供一种智能终端，上述智能终端包括存储器、处理器以及存储在上述存储器上并可在上述处理器上运行的储能系统故障检测程序，上述储能系统故障检测程序被上述处理器执行时实现上述任意一种储能系统故障检测方法的步骤。
由上可见，本发明方案中，基于预先构建的数字孪生系统获取储能系统对应的至少一组无故障数据，其中，上述无故障数据包括上述储能系统在无故障时段内的电力信息变化数据和设备部件图像数据；对上述无故障数据进行主成分分析处理，计算获得上述无故障数据对应的T 2统计量限值、Q统计量限值以及统计特征；获取上述储能系统对应的一组待检测数据，根据上述待检测数据和上述统计特征计算获得上述待检测数据对应的T 2统计量和Q统计量，其中，上述待检测数据包括上述储能系统在待检测时段内的电力信息变化数据和设备部件图像数据；根据第一大小关系和/或第二大小关系判断上述待检测数据是否为故障数据，其中，上述第一大小关系是上述T 2统计量限值与上述T 2统计量之间的大小关系，上述第二大小关系是上述Q统计量限值与上述Q统计量之间的大小关系；当上述待检测数据是故障数据时，通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别；其中，上述T 2统计量是霍特林统计量，上述Q统计量是平方预测误差统计量，上述T 2统计量限值是霍特林统计量限值，上述Q统计量限值是平方预测误差统计量限值。
有益效果
与现有技术中只能基于单一的数据指标对储能系统的故障进行判断的方案相比，本发明中一组待检测数据是结合多种数据构成的，具体包括储能系统在待检测时间段内的电力信息变化数据和设备部件图像数据。本发明中先根据储能系统对应的一组无故障数据计算获得T 2统计量限值和Q统计量限值，对于待检测数据则计算获得其T 2统计量和Q统计量，根据对应的统计量和统计量限值之间的大小关系可以先分析判断出待检测数据是否为故障数据，当其为故障数据时再进一步通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别。本发明中可以结合待检测数据中的多种信息（包括电力信息变化数据和设备部件图像数据）进行故障判断和检测，有利于提高故障检测的准确性。
附图说明
为了更清楚地说明本发明实施例中的技术方案，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性的前提下，还可以根据这些附图获得其它的附图。
图1是本发明实施例提供的一种储能系统故障检测方法的流程示意图；
图2是本发明实施例提供的一种构建数字孪生系统的流程示意图；
图3是本发明实施例提供的一种储能系统故障检测装置的结构示意图；
图4是本发明实施例提供的一种智能终端的内部结构原理框图。
 本发明的实施方式
以下描述中，为了说明而不是为了限定，提出了诸如特定系统结构、技术之类的具体细节，以便透彻理解本发明实施例。然而，本领域的技术人员应当清楚，在没有这些具体细节的其它实施例中也可以实现本发明。在其它情况下，省略对众所周知的系统、装置、电路以及方法的详细说明，以免不必要的细节妨碍本发明的描述。
应当理解，当在本说明书和所附权利要求书中使用时，术语“包括”指示所描述特征、整体、步骤、操作、元素和/或组件的存在，但并不排除一个或多个其它特征、整体、步骤、操作、元素、组件和/或其集合的存在或添加。
还应当理解，在本发明说明书中所使用的术语仅仅是出于描述特定实施例的目的而并不意在限制本发明。如在本发明说明书和所附权利要求书中所使用的那样，除非上下文清楚地指明其它情况，否则单数形式的“一”、“一个”及“该”意在包括复数形式。
还应当进一步理解，在本发明说明书和所附权利要求书中使用的术语“和/或”是指相关联列出的项中的一个或多个的任何组合以及所有可能组合，并且包括这些组合。
如在本说明书和所附权利要求书中所使用的那样，术语“如果”可以依据上下文被解释为“当…时”或“一旦”或“响应于确定”或“响应于检测到”。类似的，短语“如果确定”或“如果检测到[所描述条件或事件]”可以依据上下文被解释为意指“一旦确定”或“响应于确定”或“一旦检测到[所描述的条件或事件]”或“响应于检测到[所描述条件或事件]”。
下面结合本发明实施例的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明的一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。
在下面的描述中阐述了很多具体细节以便于充分理解本发明，但是本发明还可以采用其它不同于在此描述的其它方式来实施，本领域技术人员可以在不违背本发明内涵的情况下做类似推广，因此本发明不受下面公开的具体实施例的限制。
随着科学技术的发展，对于储能系统的研究和控制越来越受到重视。储能系统是在对储能过程进行分析时，为了确定研究对象而划出的部分物体或空间范围。它包括能量和物质的输入和输出、能量的转换和储存设备。如果储能系统出现故障，则有可能影响用户的用电体验甚至引发危险，因此需要对储能系统的故障情况进行检测，以便及时针对故障进行处理。
储能系统往往涉及多种能量、多种设备等，是随时间变换的复杂系统。而现有技术中，通常只能根据单一的数据指标对储能系统的故障情况进行判断，例如，将测量获得的电压与预设的电压阈值进行比较从而判断储能系统是否发生故障。现有技术的问题在于，只能基于单一的数据指标对储能系统的故障进行判断，无法结合多种信息进行故障检测，不利于提高故障检测的准确性。
具体的，现有技术中一次只能处理一种形式的故障数据，并且原始的训练数据（或用于训练的故障数据）少，难以进行有效的训练。同时，人机交互的效率低下，往往只有在故障发生之后才开始派遣工人去处理，且工人必须经验丰富，根据经验知道是哪一块出现故障。
为了解决上述多个问题中的至少一个问题，本发明方案中，基于预先构建的数字孪生系统获取储能系统对应的至少一组无故障数据，其中，上述无故障数据包括上述储能系统在无故障时段内的电力信息变化数据和设备部件图像数据；对上述无故障数据进行主成分分析处理，计算获得上述无故障数据对应的T 2统计量限值、Q统计量限值以及统计特征；获取上述储能系统对应的一组待检测数据，根据上述待检测数据和上述统计特征计算获得上述待检测数据对应的T 2统计量和Q统计量，其中，上述待检测数据包括上述储能系统在待检测时段内的电力信息变化数据和设备部件图像数据；根据第一大小关系和/或第二大小关系判断上述待检测数据是否为故障数据，其中，上述第一大小关系是上述T 2统计量限值与上述T 2统计量之间的大小关系，上述第二大小关系是上述Q统计量限值与上述Q统计量之间的大小关系；当上述待检测数据是故障数据时，通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别；其中，上述T 2统计量是霍特林统计量，上述Q统计量是平方预测误差统计量，上述T 2统计量限值是霍特林统计量限值，上述Q统计量限值是平方预测误差统计量限值。
与现有技术中只能基于单一的数据指标对储能系统的故障进行判断的方案相比，本发明中一组待检测数据是结合多种数据构成的，具体包括储能系统在待检测时间段内的电力信息变化数据和设备部件图像数据。本发明中先根据储能系统对应的一组无故障数据计算获得T 2统计量限值和Q统计量限值，对于待检测数据则计算获得其T 2统计量和Q统计量，根据对应的统计量和统计量限值之间的大小关系可以先分析判断出待检测数据是否为故障数据，当其为故障数据时再进一步通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别。本发明中可以结合待检测数据中的多种信息（包括电力信息变化数据和设备部件图像数据）进行故障判断和检测，有利于提高故障检测的准确性。
并且本发明中不会对于所有的待检测数据都通过故障检测模型进行故障检测，而是先判断出有故障的数据再通过故障检测模型进行进一步故障检测，可以减小计算量，并且提高故障检测的效率。对应的，故障检测模型的训练数据也可以仅为有故障的数据，可以提高故障检测模型的训练速度。
同时，本发明中还可以对采集的原始数据进行样本扩增，从而增加可用于训练的数据量。并且，构建数字孪生系统，基于数字孪生系统可以实现追踪定位，快速解决问题，并且可以进行故障可视化，以及针对故障进行模拟调整，提高在故障发生时的故障解决效率。
示例性方法
如图1所示，本发明实施例提供一种储能系统故障检测方法，具体的，上述方法包括如下步骤：
步骤S100，基于预先构建的数字孪生系统获取储能系统对应的至少一组无故障数据，其中，上述无故障数据包括上述储能系统在无故障时段内的电力信息变化数据和设备部件图像数据。
本实施例中，上述数字孪生系统是预先根据需要进行监控和故障检测的储能系统构建的。上述无故障数据是储能系统不存在故障时对应的一组数据。需要说明的是，储能系统不存在故障是指储能系统能够按照预先限定的数据指标（例如限定的电压、电流、功率等）正常运行，且储能系统中的设备不存在损坏。在一种应用场景中，上述无故障时段是任意一个不存在故障的预设时间长度的时间段；在另一种应用场景中，上述无故障时段是预先设置或者由用户选取的一个不存在故障的预设时间长度的时间段，在此不作具体限定。需要说明的是，上述无故障时段的时间长度是预先设置的，且无故障时段的时间长度与待检测时段的时间长度相同。
具体的，上述电力信息变化数据是在一段时间（如无故障时段）内的时间序列信号，即在该段时间内基于数字孪生系统进行连续多次数据采集获得的，上述设备部件图像数据则是在该段时间内的一个时刻（本实施例中为进行故障检测的当前时刻）对储能系统中的设备部件进行图像采集获得的二维视觉信号（可以是深度图像或彩色图像），需要说明的是，上述设备部件图像数据可以是基于数字孪生系统中的设备部件获得的，也可以是对实际的储能系统中的设备部件进行图像采集获得的，在此不作具体限定。本实施例中的无故障数据和待检测数据都是由电力信息变化数据和设备部件图像数据构成的，即是使用两种不同的手段获取的异构信号组成的，进而可以基于多源异构数据综合多种信息进行故障检测，提高故障检测的准确性。
本实施例中，上述电力信息变化数据包括电压数据、电流数据、温度数据和储能变化数据，上述设备部件图像数据包括上述储能系统中太阳能电池表面图像数据和风力设备叶片表面图像数据。本实施例中，上述储能系统是基于风力发电和太阳能发电的新能源储能系统，因此主要关注太阳能电池和风力设备的表面图像数据，从而便于判断是否出现太阳能电池板破裂、风机叶片出现裂痕等故障。在一种应用场景中，上述电力信息变化数据还可以包括电压差数据、储能密度数据、储能功率数据、蓄能效率数据等，在此不作具体限定。
图2是本发明实施例提供的一种构建数字孪生系统的流程示意图，如图2所示，本实施例中，在上述步骤S100之前，上述储能系统故障检测方法还包括：
步骤A100，采集获取上述储能系统对应的多组原始数据，其中，每一组上述原始数据包括上述储能系统在一个采集时段内的电力信息变化数据、设备部件图像数据以及故障标注信息，上述故障标注信息用于指示上述原始数据是故障数据或无故障数据。
步骤A200，基于上述原始数据进行样本扩增获得多组扩增数据，其中，对上述电力信息变化数据和上述设备部件图像数据采用不同的扩增通道进行样本扩增。
步骤A300，将上述原始数据和上述扩增数据上传到数字孪生数据库，并构建与上述储能系统对应的数字孪生系统。
数字孪生是充分利用物理模型、传感器更新、运行历史等数据，集成多学科、多物理量、多尺度、多概率的仿真过程，本实施例中的数字孪生系统是上述储能系统对应的数字映射系统。进一步的，本实施例中，数字孪生系统是根据储能系统中的设备、设备之间的关系、设备参数、设备运行状态等构建的，数字孪生系统还可以与储能系统进行数据交互。
本实施例中，采集获得多组原始数据，每一组原始数据可以是存在故障的数据，也可以是不存在故障的数据，根据原始数据中的故障标注信息进行区分确定。每一组原始数据在不同的采集时段内采集获得，但采集时段的具体时间长度是相同的，且采集时段的具体时间长度与上述无故障时段的具体时间长度相同。
本实施例中，还基于原始数据进行样本扩增，以增加数字孪生系统中的数据量。在一种应用场景中，对于每一组原始数据可以进行样本扩增获得一组或多组对应的扩增数据。同时，电力信息变化数据是时间序列数据，而设备部件图像数据是图像数据，两者有不同的特点，因此本实施例中采用不同的方式分别对这两种数据进行扩增，以获得更好的数据扩增结果。
本实施例中，采用双通道模式对原始数据进行扩增，以解决数据量不足、尤其是故障发生随机的问题。对于电力信息变化数据，采用生成对抗网络（GAN，Generative Adversarial Network）进行扩增处理，对于设备部件图像数据，采用深度卷积生成对抗网络（DCGAN，Deep Convolution Generative Adversarial Networks）进行扩增处理。同时，也可以基于样本扩增获得更多的故障数据，有利于为故障检测模型的训练过程提供更多训练数据。
需要说明的是，扩增数据中包括扩增获得的电力信息变化数据、设备部件图像数据以及对应的故障标注信息。在一种应用场景中，基于双通道通过不同方式拟合获得扩增数据（例如针对图像进行平滑处理和拟合，针对时间序列信号进行曲线拟合），此时扩增数据的故障标注信息与对应的原始数据相同。在另一种应用场景中，基于双通道通过不同方式处理原始数据，并添加故障噪声，此时可以获得扩增的故障数据。
对于已经生成的扩增数据与原始数据，全部上传到数字孪生数据库中，在虚拟世界中生成与现实世界一样的储能系统。
步骤S200，对上述无故障数据进行主成分分析处理，计算获得上述无故障数据对应的T 2统计量限值、Q统计量限值以及统计特征。
需要说明的是，在一种应用场景中，对上述无故障数据进行双通道的主成分分析，即分别对其中的电力信息变化数据和设备部件图像数据进行主成分分析处理，分别获得电力信息变化数据和设备部件图像数据对应的T 2统计量限值、Q统计量限值以及统计特征。在另一种应用场景中，也可以将设备部件图像数据处理成与电力信息变化数据形式相同的数值数据，从而与电力信息变化数据一起进行主成分分析处理，在此不作具体限定。
具体的，可以基于获取的一组或多组无故障数据计算对应的统计量限值，本实施例中以对一组无故障数据进行计算为例进行说明，当有多组无故障数据时，可以分别针对每一组无故障数据计算对应的T 2统计量限值、Q统计量限值以及统计特征，并根据计算出的两组统计特征分别执行之后的计算T 2统计量和Q统计量的步骤，最后综合多组大小关系确定待检测数据是否为故障数据。
本实施例中，以对一种数据进行主成分分析处理为例进行说明，但不作为具体限定。本实施例中，基于主成分分析（PCA，principal components analysis）技术进行降维处理，基于PCA的故障诊断过程中，通过统计量限值和对应的统计量的大小关系可以判断对应的数据是否存在故障。
需要说明的是，具体的统计特征并不作限定，例如，一种应用场景中，统计特征可以包括均值和标准差，此时在对待检测数据进行处理时基于均值和标准差将待检测数据中的各个特征数据进行标准化。在另一种应用场景中，统计特征可以包括均值和均方根，此时在对待检测数据进行处理时基于均值和均方根将待检测数据中的各个特征数据进行标准化。
本实施例中，以统计特征包括均值和均方根为例进行说明，具体包括上述无故障数据中每一个特征（即每一种电力信息变化数据和每一种设备部件图像数据）的均值和标准差。
进一步的，也可以先对上述无故障数据进行预处理，对其进行归一化，针对其中的每个特征归一化均值为0，均方根为1的数据，以减小计算量，提高处理效率。
在一种应用场景中，无故障数据为n行m列的矩阵X * n*m，其中每一个元素用X * ij代表，i是1到n的整数，j是1到m的整数，n代表样本个数，m代表样本维数（即特征个数）。本实施例中，将无故障数据中的每一个特征归一化为均值为0、均方根为1的数据，获得归一化无故障数据。
对上述归一化无故障数据进行PCA降维并获得对应的k个特征。具体的，计算获得上述归一化无故障数据对应的矩阵的协方差矩阵，求取协方差矩阵的特征值和特征向量，并将特征值按照从大到小的顺序排列。在一种应用场景中，特征值按照从小到大的顺序排列为λ 1、λ 2、λ 3…λ m，对应的特征向量按照从大到小的顺序排列后获得P mm=[p 1,p 2,…p m]，k为选取的特征的数目（即主元个数），k的值可以预先设定，也可以根据实际需求选择。例如，可以预先设置k为3，或者，可以选择特征值累计大于预设比例阈值（例如70%）的前k个特征进行PCA降维。降维以后得到的矩阵中样本个数仍然为n，但样本维数降低为k。
进一步的，根据选取的特征的数目以及预先设置的统计量限值计算公式分别计算获得T 2统计量限值和Q统计量限值，具体的统计量限值计算公式和计算过程可以参照现有技术中PCA降维过程中对应统计量限值的求解公式和求解过程，在此不作具体限定。
步骤S300，获取上述储能系统对应的一组待检测数据，根据上述待检测数据和上述统计特征计算获得上述待检测数据对应的T 2统计量和Q统计量，其中，上述待检测数据包括上述储能系统在待检测时段内的电力信息变化数据和设备部件图像数据。
本实施例中，上述待检测数据是在储能系统运行过程中对储能系统进行数据采集获得的一组数据。在一种应用场景中，上述待检测数据还可以是对数字孪生系统进行某一操作之后，获得的一段时间内的系统运行预测数据（以预测这个操作是否会引起故障），在此不作具体限定。待检测数据的时间长度与无故障数据的时间长度相同，且数据形式也相同，检测时段的时间长度与无故障时段的时间长度相同。
上述根据上述待检测数据和上述统计特征计算获得上述待检测数据对应的T 2统计量和Q统计量，包括：根据上述统计特征对上述待检测数据进行标准化处理，获得待检测标准数据；根据上述待检测标准数据计算获取上述待检测数据对应的T 2统计量和Q统计量。
需要说明的是，本实施例中，上述无故障数据先进行了标准化，因此对应的统计特征中均值为0，均方根为1，据此对待检测数据进行标准化处理，即将待检测数据的均值和均方根也处理为0和1，获得对应的待检测标准数据。
其中，上述T 2统计量和Q统计量可以根据预先设置的统计量计算公式计算获得，具体的统计量计算公式和计算过程可以参照现有技术中PCA降维过程中对应统计量的求解公式和求解过程，在此不作具体限定。需要说明的是，在T 2统计量和Q统计量计算过程中，使用的主元个数与上述计算统计量限值时使用的主元个数k相同。
步骤S400，根据第一大小关系和/或第二大小关系判断上述待检测数据是否为故障数据。
其中，上述第一大小关系是上述T 2统计量限值与上述T 2统计量之间的大小关系，上述第二大小关系是上述Q统计量限值与上述Q统计量之间的大小关系。
具体的，当上述T 2统计量大于上述T 2统计量限值和/或上述Q统计量大于上述Q统计量限值时，上述待检测数据为故障数据，否则上述待检测数据不是故障数据。
在一种应用场景中，当上述T 2统计量与预先设置的第一统计量补充值之和大于上述T 2统计量限值和/或上述Q统计量与预先设置的第二统计量补充值之和大于上述Q统计量限值时，上述待检测数据为故障数据，否则上述待检测数据不是故障数据。上述第一统计量补充值和第二统计量补充值是可以根据实际需求预先设置和调整的。
本实施例中先使用PCA进行故障检测，用正常数据训练PCA获得T 2统计量限值和Q统计量限值，使用待检测数据进行测试的时候，得到待检测数据的T 2统计量和Q统计量，如果某一个统计量超过对应的统计量阈值，则认为该待检测数据发生了故障，此时会进一步进行故障类别的检测。
步骤S500，当上述待检测数据是故障数据时，通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别。
其中，上述T 2统计量是霍特林统计量，上述Q统计量是平方预测误差（SPE，Squared prediction error）统计量，上述T 2统计量限值是霍特林统计量限值，上述Q统计量限值是平方预测误差统计量限值。霍特林统计量、平方预测误差统计量、霍特林统计量限值以及平方预测误差统计量限值是在进行主成分分析处理过程中计算获得的统计量或限值，其具体计算公式和计算过程可以参照现有技术中PCA降维过程中对应统计量的求解公式和求解过程，在此不作具体限定。
具体的，上述已训练的故障检测模型是预先训练好的故障检测模型，本实施例中，上述待检测数据中包括两种不同类型的数据，因此上述故障检测模型可以使用两个不同的通道分别对其进行处理，实现双通道的故障分类。
上述步骤S500具体包括：将上述待检测数据输入上述已训练的故障检测模型，通过上述已训练的故障检测模型输出上述待检测数据对应的故障类别，其中，上述已训练的故障检测模型基于逻辑回归方法对上述待检测数据中的电力信息变化数据进行故障分类，上述已训练的故障检测模型基于VGG-Net对上述待检测数据中的设备部件图像数据进行故障分类。本实施例中，上述两个部分（包括逻辑回归部分和VGG-Net部分）设置在同一个模型中，同时进行训练，如此可以充分利用两种数据之间的关联性进行训练，提高训练后的故障检测模型的准确性。
在一种应用场景中，上述故障检测模型中可以包括两个子模型，分别处理电力信息变化数据和设备部件图像数据。在另一种应用场景中，也可以使用两个单独训练的子模型构成上述故障检测模型，但不作为具体限定。
如此，对于发生了故障的数据进行故障分类，对于时间序列使用逻辑回归的方法进行判断是哪一类故障。对于图片数据使用VGG-net 进行故障分类，判断是哪一类故障。
其中，上述故障类别可以根据实际需求预先设置，例如，故障类别可以包括电压不稳定、电流过高、太阳能电池板破裂、风力叶片破裂、其它某一具体器件或设备损坏等，在此不作具体限定。
进一步的，上述已训练的故障检测模型根据如下步骤进行训练：
将训练数据中的电力信息训练数据和设备部件图像训练数据输入上述故障检测模型，通过上述故障检测模型对上述电力信息训练数据和上述设备部件图像训练数据进行故障分类并获得检测故障类别，其中，上述训练数据包括多组训练信息组，每一组上述训练信息组包括电力信息训练数据、设备部件图像训练数据以及标注故障类别；
根据上述检测故障类别和对应的上述标注故障类别对上述故障检测模型的模型参数进行调整，并继续执行上述将训练数据中的电力信息训练数据和设备部件图像训练数据输入上述故障检测模型的步骤，直至满足预设训练条件，以得到已训练的故障检测模型。
本实施例中，上述故障检测模型只需要对存在故障的数据进行类别检测，因此上述训练数据可以仅包括有故障的数据，从而降低计算量。
需要说明的是，本实施例中，上述故障检测模型可以分别对两种数据（电力信息训练数据和设备部件图像训练数据）进行处理，因此两种数据也可以分别有对应的故障类别，例如，上述检测故障类别可以包括第一检测故障类别和第二检测故障类别，上述标注故障类别可以包括第一标注故障类别和第二标注故障类别，第一检测故障类别和第一标注故障类别对应，第二检测故障类别与第二标注故障类别对应。如此，在根据检测故障类别和标注故障类别对故障检测模型的模型参数进行调整时，可以分别根据第一检测故障类别和第一标注故障类别、第二检测故障类别与第二标注故障类别计算对应的损失值，根据损失值进行模型参数的调整。需要说明的是，还可以为两种不同数据对应的损失设置不同的阈值，灵活调整两种数据的受关注程度。
其中，上述预设的训练条件为训练时的迭代次数达到预设的迭代阈值，或者计算获得的损失值达到小于预设的损失阈值。
可选的，在上述通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别之后，上述方法还包括：基于上述故障类别在上述数字孪生系统中进行故障标记并将标记结果进行可视化输出。具体的，在本实施例中，结合数字孪生系统能同时处理储能系统中存在的时间序列和图片两种不同形式的故障数据。并对数据进行故障检测，分类并实时上传到数字孪生的数据库中，实时在虚拟世界中显示有没有故障，有故障的话是哪一种故障。还可以根据检测出的故障类别和预设的故障类别与损坏部件之间的关系（例如预先建立的故障类别与损坏部件关系表）确定当前是哪些部件存在对应的故障，将这些部件在数字孪生系统中进行标记并进行可视化输出，实现追踪定位，方便用户及时查看和确认故障发生的位置，以便进行进一步维修。进一步的，在上述通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别之后，上述方法还包括：基于上述故障类别获取故障调整策略；将上述故障调整策略输入上述数字孪生系统，并在上述数字孪生系统中执行上述故障调整策略；获取上述数字孪生系统中的故障调整结果并输出。其中，上述故障调整策略可以由目标对象（即操作人员）输入，也可以根据上述故障类别在预先设置的调整策略表中搜索获得，还以在云端搜索获得，在此不作具体限定。上述故障调整策略是针对故障类别对于储能系统中的设备进行调整的具体方案，获得故障调整策略之后，输入数字孪生系统，如此，可以在数字孪生系统中先进行策略的模拟执行，获得故障调整结果。进一步的，如果故障调整结果为调整有效（即故障消除或减弱，例如对应的指标恢复正常）则可以在实际的储能系统中也执行对应的策略，反之则重新获取新的故障调整策略并在数字孪生系统中重新执行，如此，基于数字孪生系统可以有效提高故障调整的效率，降低处理难度，且避免误操作。
由上可见，本实施例中，一组待检测数据是结合多种数据构成的，具体包括储能系统在待检测时间段内的电力信息变化数据和设备部件图像数据。本实施例中先根据储能系统对应的一组无故障数据计算获得T 2统计量限值和Q统计量限值，对于待检测数据则计算获得其T 2统计量和Q统计量，根据对应的统计量和统计量限值之间的大小关系可以先分析判断出待检测数据是否为故障数据，当其为故障数据时再进一步通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别。本实施例中可以结合待检测数据中的多种信息（包括电力信息变化数据和设备部件图像数据）进行故障判断和检测，有利于提高故障检测的准确性。
并且本实施例中不会对于所有的待检测数据都通过故障检测模型进行故障检测，而是先判断出有故障的数据再通过故障检测模型进行进一步故障检测，可以减小计算量，并且提高故障检测的效率。对应的，故障检测模型的训练数据也可以仅为有故障的数据，可以提高故障检测模型的训练速度。
同时，本实施例中还可以对采集的原始数据进行样本扩增，从而增加可用于训练的数据量。并且，构建数字孪生系统，基于数字孪生系统可以实现追踪定位，快速解决问题，并且可以进行故障可视化，以及针对故障进行模拟调整，提高在故障发生时的故障解决效率。
示例性设备
如图3中所示，对应于上述储能系统故障检测方法，本发明实施例还提供一种储能系统故障检测装置，上述储能系统故障检测装置包括：
无故障数据获取模块610，用于基于预先构建的数字孪生系统获取储能系统对应的至少一组无故障数据，其中，上述无故障数据包括上述储能系统在无故障时段内的电力信息变化数据和设备部件图像数据。
统计量限值计算模块620，用于对上述无故障数据进行主成分分析处理，计算获得上述无故障数据对应的T 2统计量限值、Q统计量限值以及统计特征。
待检测数据处理模块630，用于获取上述储能系统对应的一组待检测数据，根据上述待检测数据和上述统计特征计算获得上述待检测数据对应的T 2统计量和Q统计量，其中，上述待检测数据包括上述储能系统在待检测时段内的电力信息变化数据和设备部件图像数据。
故障数据判断模块640，用于根据第一大小关系和/或第二大小关系判断上述待检测数据是否为故障数据，其中，上述第一大小关系是上述T 2统计量限值与上述T 2统计量之间的大小关系，上述第二大小关系是上述Q统计量限值与上述Q统计量之间的大小关系。
故障类别检测模块650，用于当上述待检测数据是故障数据时，通过已训练的故障检测模型对上述待检测数据进行故障检测并获得故障类别。
其中，上述统计量是霍特林统计量，上述Q统计量是平方预测误差（SPE，Squared prediction error）统计量，上述T 2统计量限值是霍特林统计量限值，上述Q统计量限值是平方预测误差统计量限值。霍特林统计量、平方预测误差统计量、霍特林统计量限值以及平方预测误差统计量限值是在进行主成分分析处理过程中计算获得的统计量或限值，其具体计算公式和计算过程可以参照现有技术中PCA降维过程中对应统计量的求解公式和求解过程，在此不作具体限定。
具体的，本实施例中，上述储能系统故障检测装置及其各模块的具体功能可以参照上述储能系统故障检测方法中的对应描述，在此不再赘述。
需要说明的是，上述储能系统故障检测装置的各个模块的划分方式并不唯一，在此也不作为具体限定。
基于上述实施例，本发明还提供了一种智能终端，其原理框图可以如图4所示。上述智能终端包括处理器及存储器。该智能终端的存储器包括储能系统故障检测程序，存储器为储能系统故障检测程序的运行提供环境。该储能系统故障检测程序被处理器执行时实现上述任意一种储能系统故障检测方法的步骤。需要说明的是，上述智能终端还可以包括其它功能模块或单元，在此不作具体限定。
本领域技术人员可以理解，图4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智能终端的限定，具体地智能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。
本发明实施例还提供一种计算机可读存储介质，上述计算机可读存储介质上存储有储能系统故障检测程序，上述储能系统故障检测程序被处理器执行时实现本发明实施例提供的任意一种储能系统故障检测方法的步骤。
应理解，上述实施例中各步骤的序号大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本发明实施例的实施过程构成任何限定。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将上述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。实施例中的各功能单元、模块可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中，上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。另外，各功能单元、模块的具体名称也只是为了便于相互区分，并不用于限制本发明的保护范围。上述装置中单元、模块的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述或记载的部分，可以参见其它实施例的相关描述。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各实例的单元及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟是以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。
在本发明所提供的实施例中，应该理解到，所揭露的装置/智能终端和方法，可以通过其它的方式实现。例如，以上所描述的装置/智能终端实施例仅仅是示意性的，例如，上述模块或单元的划分，仅仅为一种逻辑功能划分，实际实现时可以由另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个装置，或一些特征可以忽略，或不执行。
上述集成的模块/单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读存储介质中。基于这样的理解，本发明实现上述实施例方法中的全部或部分流程，也可以通过计算机程序来指令相关的硬件来完成，上述计算机程序可存储于一计算机可读存储介质中，该计算机程序在被处理器执行时，可实现上述各个方法实施例的步骤。其中，上述计算机程序包括计算机程序代码，上述计算机程序代码可以为源代码形式、对象代码形式、可执行文件或某些中间形式等。上述计算机可读介质可以包括：能够携带上述计算机程序代码的任何实体或装置、记录介质、U盘、移动硬盘、磁碟、光盘、计算机存储器、只读存储器（ROM，Read-Only Memory）、随机存取存储器（RAM，Random Access Memory）、电载波信号、电信信号以及软件分发介质等。需要说明的是，上述计算机可读存储介质包含的内容可以根据司法管辖区内立法和专利实践的要求进行适当的增减。
以上所述实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解；其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不是相应技术方案的本质脱离本发明各实施例技术方案的精神和范围，均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种储能系统故障检测方法，其特征在于，所述方法包括：
    基于预先构建的数字孪生系统获取储能系统对应的至少一组无故障数据，其中，所述无故障数据包括所述储能系统在无故障时段内的电力信息变化数据和设备部件图像数据；
    对所述无故障数据进行主成分分析处理，计算获得所述无故障数据对应的T 2统计量限值、Q统计量限值以及统计特征；
    获取所述储能系统对应的一组待检测数据，根据所述待检测数据和所述统计特征计算获得所述待检测数据对应的T 2统计量和Q统计量，其中，所述待检测数据包括所述储能系统在待检测时段内的电力信息变化数据和设备部件图像数据；
    根据第一大小关系和/或第二大小关系判断所述待检测数据是否为故障数据，其中，所述第一大小关系是所述T 2统计量限值与所述T 2统计量之间的大小关系，所述第二大小关系是所述Q统计量限值与所述Q统计量之间的大小关系；
    当所述待检测数据是故障数据时，通过已训练的故障检测模型对所述待检测数据进行故障检测并获得故障类别；
    其中，所述T 2统计量是霍特林统计量，所述Q统计量是平方预测误差统计量，所述T 2统计量限值是霍特林统计量限值，所述Q统计量限值是平方预测误差统计量限值。
  2. 根据权利要求1所述的储能系统故障检测方法，其特征在于，所述电力信息变化数据包括电压数据、电流数据、温度数据和储能变化数据，所述设备部件图像数据包括所述储能系统中太阳能电池表面图像数据和风力设备叶片表面图像数据。
  3. 根据权利要求1所述的储能系统故障检测方法，其特征在于，在所述基于预先构建的数字孪生系统获取储能系统对应的至少一组无故障数据之前，所述方法还包括：
    采集获取所述储能系统对应的多组原始数据，其中，每一组所述原始数据包括所述储能系统在一个采集时段内的电力信息变化数据、设备部件图像数据以及故障标注信息，所述故障标注信息用于指示所述原始数据是故障数据或无故障数据；
    基于所述原始数据进行样本扩增获得多组扩增数据，其中，对所述电力信息变化数据和所述设备部件图像数据采用不同的扩增通道进行样本扩增；
    将所述原始数据和所述扩增数据上传到数字孪生数据库，并构建与所述储能系统对应的数字孪生系统。
  4. 根据权利要求1所述的储能系统故障检测方法，其特征在于，所述统计特征包括均值和标准差，所述根据所述待检测数据和所述统计特征计算获得所述待检测数据对应的T 2统计量和Q统计量，包括：
    根据所述统计特征对所述待检测数据进行标准化处理，获得待检测标准数据；
    根据所述待检测标准数据计算获取所述待检测数据对应的T 2统计量和Q统计量。
  5. 根据权利要求1所述的储能系统故障检测方法，其特征在于，所述当所述待检测数据是故障数据时，通过已训练的故障检测模型对所述待检测数据进行故障检测并获得故障类别，包括：
    将所述待检测数据输入所述已训练的故障检测模型，通过所述已训练的故障检测模型输出所述待检测数据对应的故障类别，其中，所述已训练的故障检测模型基于逻辑回归方法对所述待检测数据中的电力信息变化数据进行故障分类，所述已训练的故障检测模型基于VGG-Net对所述待检测数据中的设备部件图像数据进行故障分类。
  6. 根据权利要求1所述的储能系统故障检测方法，其特征在于，所述已训练的故障检测模型根据如下步骤进行训练：
    将训练数据中的电力信息训练数据和设备部件图像训练数据输入所述故障检测模型，通过所述故障检测模型对所述电力信息训练数据和所述设备部件图像训练数据进行故障分类并获得检测故障类别，其中，所述训练数据包括多组训练信息组，每一组所述训练信息组包括电力信息训练数据、设备部件图像训练数据以及标注故障类别；
    根据所述检测故障类别和对应的所述标注故障类别对所述故障检测模型的模型参数进行调整，并继续执行所述将训练数据中的电力信息训练数据和设备部件图像训练数据输入所述故障检测模型的步骤，直至满足预设训练条件，以得到已训练的故障检测模型。
  7. 根据权利要求1-6任意一项所述的储能系统故障检测方法，其特征在于，在所述通过已训练的故障检测模型对所述待检测数据进行故障检测并获得故障类别之后，所述方法还包括：
    基于所述故障类别在所述数字孪生系统中进行故障标记并将标记结果进行可视化输出。
  8. 根据权利要求1-6任意一项所述的储能系统故障检测方法，其特征在于，在所述通过已训练的故障检测模型对所述待检测数据进行故障检测并获得故障类别之后，所述方法还包括：
    基于所述故障类别获取故障调整策略；
    将所述故障调整策略输入所述数字孪生系统，并在所述数字孪生系统中执行所述故障调整策略；
    获取所述数字孪生系统中的故障调整结果并输出。
  9. 一种储能系统故障检测装置，其特征在于，所述装置包括：
    无故障数据获取模块，用于基于预先构建的数字孪生系统获取储能系统对应的至少一组无故障数据，其中，所述无故障数据包括所述储能系统在无故障时段内的电力信息变化数据和设备部件图像数据；
    统计量限值计算模块，用于对所述无故障数据进行主成分分析处理，计算获得所述无故障数据对应的T 2统计量限值、Q统计量限值以及统计特征；
    待检测数据处理模块，用于获取所述储能系统对应的一组待检测数据，根据所述待检测数据和所述统计特征计算获得所述待检测数据对应的T 2统计量和Q统计量，其中，所述待检测数据包括所述储能系统在待检测时段内的电力信息变化数据和设备部件图像数据；
    故障数据判断模块，用于根据第一大小关系和/或第二大小关系判断所述待检测数据是否为故障数据，其中，所述第一大小关系是所述T 2统计量限值与所述T 2统计量之间的大小关系，所述第二大小关系是所述Q统计量限值与所述Q统计量之间的大小关系；
    故障类别检测模块，用于当所述待检测数据是故障数据时，通过已训练的故障检测模型对所述待检测数据进行故障检测并获得故障类别；
    其中，所述T 2统计量是霍特林统计量，所述Q统计量是平方预测误差统计量，所述T 2统计量限值是霍特林统计量限值，所述Q统计量限值是平方预测误差统计量限值。
  10. 一种智能终端，其特征在于，所述智能终端包括存储器、处理器以及存储在所述存储器上并可在所述处理器上运行的储能系统故障检测程序，所述储能系统故障检测程序被所述处理器执行时实现如权利要求1-8任意一项所述储能系统故障检测方法的步骤。
PCT/CN2022/136761 2022-10-08 2022-12-06 储能系统故障检测方法、装置及智能终端 WO2024073932A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211219619.2 2022-10-08
CN202211219619.2A CN116150666B (zh) 2022-10-08 2022-10-08 储能系统故障检测方法、装置及智能终端

Publications (1)

Publication Number Publication Date
WO2024073932A1 true WO2024073932A1 (zh) 2024-04-11

Family

ID=86353212

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/136761 WO2024073932A1 (zh) 2022-10-08 2022-12-06 储能系统故障检测方法、装置及智能终端

Country Status (2)

Country Link
CN (1) CN116150666B (zh)
WO (1) WO2024073932A1 (zh)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150074469A1 (en) * 2013-09-09 2015-03-12 International Business Machines Corporation Methods, apparatus and system for notification of predictable memory failure
WO2018028005A1 (zh) * 2016-08-11 2018-02-15 苏州瑞得恩自动化设备科技有限公司 一种大型光伏电站中电池面板的故障检测算法
CN109001649A (zh) * 2018-07-21 2018-12-14 成都光电传感技术研究所有限公司 一种电源智能诊断系统及保护方法
CN111913475A (zh) * 2020-08-07 2020-11-10 清华大学 动态系统中故障检测方法、检测器以及存储介质
CN115034471A (zh) * 2022-06-09 2022-09-09 天津蓝天太阳科技有限公司 一种储能故障预测方法及其装置

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11237551B2 (en) * 2018-07-13 2022-02-01 United States Of America, As Represented By The Secretary Of The Navy Real-time automation for monitor and control of electro-mechanical system
EP3923309A1 (en) * 2020-06-12 2021-12-15 ABB Power Grids Switzerland AG Provision of currently unmeasurable state information about a switchgear device system
CN112731159B (zh) * 2020-12-23 2023-09-22 江苏省电力试验研究院有限公司 一种储能电站电池舱电池故障预判及定位的方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150074469A1 (en) * 2013-09-09 2015-03-12 International Business Machines Corporation Methods, apparatus and system for notification of predictable memory failure
WO2018028005A1 (zh) * 2016-08-11 2018-02-15 苏州瑞得恩自动化设备科技有限公司 一种大型光伏电站中电池面板的故障检测算法
CN109001649A (zh) * 2018-07-21 2018-12-14 成都光电传感技术研究所有限公司 一种电源智能诊断系统及保护方法
CN111913475A (zh) * 2020-08-07 2020-11-10 清华大学 动态系统中故障检测方法、检测器以及存储介质
CN115034471A (zh) * 2022-06-09 2022-09-09 天津蓝天太阳科技有限公司 一种储能故障预测方法及其装置

Also Published As

Publication number Publication date
CN116150666B (zh) 2023-10-31
CN116150666A (zh) 2023-05-23

Similar Documents

Publication Publication Date Title
US8868985B2 (en) Supervised fault learning using rule-generated samples for machine condition monitoring
CN108375476B (zh) 一种水电机组健康评估方法
CN110636066B (zh) 基于无监督生成推理的网络安全威胁态势评估方法
CN114723285B (zh) 一种电网设备安全性评估预测方法
CN110530650B (zh) 基于广义回归神经网络与箱型图分析的重型燃气轮机性能状态监测方法
CN114619292B (zh) 一种基于小波降噪和注意力机制融合gru网络的铣削加工刀具磨损监测方法
CN116028887B (zh) 一种连续性工业生产数据的分析方法
CN111125895A (zh) 一种风力发电机整机服役功率特性在线评估方法
CN112906764A (zh) 基于改进bp神经网络的通信安全设备智能诊断方法及系统
CN111273125A (zh) 基于rst-cnn的电力电缆沟道故障诊断方法
CN115373370A (zh) 一种可编程控制器运行状态监测方法及系统
CN113487086A (zh) 设备剩余使用寿命预测方法、装置、计算机设备和介质
CN113689072A (zh) 一种基于马尔科夫链的海上风电机组运行状态评估方法
KR102523701B1 (ko) 인공지능을 이용한 태양광 모듈의 품질 예측 장치 및 방법
WO2024073932A1 (zh) 储能系统故障检测方法、装置及智能终端
CN111953657A (zh) 一种序列-数据联合驱动的配电网cps网络攻击辨识方法
CN116756505A (zh) 一种基于大数据的光伏设备智能管理系统及方法
CN108960332A (zh) 一种基于多向主元素分析法的在线监测方法
CN115828165B (zh) 一种新能源智能微电网数据处理方法及系统
CN113298415B (zh) 一种用于能量枢纽的协同运行质量分析评估方法
CN115983682A (zh) 一种虚实结合的化工生产全过程能耗监测与诊断分析方法
CN115168802A (zh) 一种基于cva和plsr的风机故障诊断方法
CN117909780A (zh) 基于智能传感设备的风电运维管理方法及系统
CN115421467A (zh) 基于多块概率核偏最小二乘的大规模工业过程故障诊断方法
CN114298495A (zh) 基于细粒度数据驱动的电网可靠性评估方法