Fault detection method, system and device and storage medium

Abstract

The invention discloses a fault detection method, system and device and a storage medium, and the method comprises the steps: obtaining the real-time data of a system, and determining a first priority data set and a second priority data set based on the real-time data of the system; inputting the first priority data set into a trained rapid fault detection model for calculation to obtain a first fault prediction probability; and judging a relationship between the first fault prediction probability and a preset threshold value, and determining whether fault prediction needs to be performed according to the second priority data set according to a judgment result. The fault detection efficiency of the power distribution calculation system can be improved, and the method can be widely applied to the technical field of distribution network power distribution.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network power distribution, and particularly to a fault detection method, system, device and storage medium. Background Art

[0002] As a key component of the power network, the stability and reliability of the distribution system directly affect people's daily life, industrial production and the overall operation of society. Once a fault occurs in the distribution system, if the problem cannot be quickly and accurately located and solved, it will lead to an expansion of the power outage range and an extension of the duration, thereby causing huge economic losses and social impacts; therefore, it is very important to detect faults by analyzing the data of the distribution system to determine whether the distribution system is abnormal.

[0003] The fault detection of the distribution system data is to analyze and compare a large amount of real-time operation data, use algorithms to identify data anomalies, and quickly locate possible fault points or performance degradation situations in the distribution system, including voltage fluctuations, current anomalies, equipment overheating, etc., so as to provide accurate diagnostic information and maintenance guidance for operation and maintenance personnel, thereby ensuring the efficient and safe operation of the distribution system.

[0004] However, in the prior art, when the distribution calculation system faces high-frequency data collection and large-scale distribution network nodes, there will be a problem of low fault detection efficiency. Summary of the Invention

[0005] In view of this, to solve one of the above problems, an object of the embodiments of the present invention is to provide a fault detection method, system, device and storage medium, which can improve the fault detection efficiency of the distribution calculation system.

[0006] In a first aspect, an embodiment of the present invention provides a fault detection method, including:

[0007] Obtain the real-time data of the system, and determine a first priority data set and a second priority data set based on the real-time data of the system;

[0008] Input the first priority data set into the trained fast fault detection model for calculation, and obtain a first fault prediction probability;

[0009] Judge the relationship between the first fault prediction probability and a preset threshold, and determine whether it is necessary to perform fault prediction according to the second priority data set according to the judgment result.

[0010] Specifically, the determining the first priority data set and the second priority data set based on the real-time data includes:

[0011] Classify the real-time data of the system based on the importance of the data to obtain first priority data and second priority data;

[0012] Extract features from the first-priority data using a first preset method to obtain first-priority data features;

[0013] Extract features from the second-priority data using a second preset method to obtain second-priority data features;

[0014] Mark the features of the first-priority data using a third preset method to construct the first-priority data set;

[0015] Mark the features of the second-priority data using a fourth preset method to construct the second-priority data set.

[0016] Specifically, the extracting features from the first-priority data using a first preset method to obtain first-priority data features; extracting features from the second-priority data using a second preset method to obtain second-priority data features includes:

[0017] Perform synchronous feature extraction on the first-priority data to obtain first-priority data features;

[0018] Perform batch or asynchronous extraction on the second-priority data to obtain second-priority data features.

[0019] Specifically, the marking the features of the first-priority data using a third preset method to construct the first-priority data set; marking the features of the second-priority data using a fourth preset method to construct the second-priority data set includes:

[0020] Perform real-time marking based on the first-priority data features to construct the first-priority data set;

[0021] Perform batch processing marking based on the second-priority data features to construct the second-priority data set.

[0022] Specifically, the fast fault detection model is trained by the following method:

[0023] Obtain a first sample data set; the first sample data set includes a first-priority data sample set and a first fault detection result sample;

[0024] Input the first-priority data sample set into a pre-fast fault detection model for calculation to obtain a first sample training result;

[0025] Update the pre-fast fault detection model based on the first fault detection result sample and the first sample training result, and obtain the fast fault detection model according to the update result.

[0026] Specifically, determining whether to perform fault prediction based on the second priority data set according to the judgment result includes:

[0027] If the judgment result is that the first fault prediction probability is greater than the preset threshold, determine that the fault detection result of the system is a fault;

[0028] If the judgment result is that the first fault prediction probability is less than the preset threshold, input the first priority data set and the second priority data set into the trained comprehensive fault detection model for calculation, and determine the fault detection result of the system according to the calculation result.

[0029] Specifically, the comprehensive fault detection model is trained by the following method:

[0030] Obtain a comprehensive sample data set; the comprehensive sample data set includes a first priority data sample set, a first fault detection result sample, a second priority data sample set, and a second fault detection result sample;

[0031] Input the first priority data sample set and the second priority data sample set into the pre-comprehensive fault detection model for calculation to obtain a comprehensive sample training result;

[0032] Based on the first fault detection result sample, the second fault detection result sample, and the comprehensive sample training result, update the pre-comprehensive fault detection model, and obtain a comprehensive fault prediction model according to the update result.

[0033] On the other hand, an embodiment of the present invention further provides a fault detection system, including:

[0034] A first module, configured to obtain real-time data of the system, and determine a first priority data set and a second priority data set based on the real-time data of the system;

[0035] A second module, configured to input the first priority data set into the trained fast fault detection model for calculation to obtain a first fault prediction probability;

[0036] A third module, configured to judge the relationship between the first fault prediction probability and a preset threshold, and determine whether to perform fault prediction according to the second priority data set according to the judgment result.

[0037] On the other hand, an embodiment of the present invention further provides a fault detection device, including:

[0038] At least one processor;

[0039] At least one memory, configured to store at least one program;

[0040] When the at least one program is executed by the at least one processor, the at least one processor implements the method as described above.

[0041] On the other hand, an embodiment of the present invention further provides a computer-readable storage medium, in which a program executable by a processor is stored, and the program executable by the processor is used to execute the method as described above when executed by the processor.

[0042] In summary, implementing the embodiments of the present invention includes the following beneficial effects:

[0043] This embodiment provides a fault detection method, system, device and storage medium. The method performs hierarchical processing on the acquired real-time system data, and preferentially performs fault detection processing on high-priority data; when the prediction result of the high-priority data of the system is a fault, the fault detection process ends, so the amount of fault detection data can be reduced, thereby improving the fault detection efficiency of the system; further, when it is determined that the high-priority data of the system does not show abnormalities, low-priority data can be combined for further fault detection. Therefore, the present invention can improve the fault detection efficiency of the system while ensuring the reliability of the system fault detection result. Description of the Drawings

[0044] Figure 1 is a schematic flow chart of the steps of a fault prediction method provided by an embodiment of the present invention;

[0045] Figure 2 is a structural block diagram of a fault prediction system provided by an embodiment of the present invention;

[0046] Figure 3 is a structural block diagram of a fault prediction device provided by an embodiment of the present invention. Detailed Embodiments

[0047] The following further describes the present invention in detail with reference to the drawings and specific embodiments. For the step numbers in the following embodiments, they are only set for the convenience of explanation and illustration, and no limitation is imposed on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0048] Explanations of several terms involved in this application are as follows:

[0049] GBDT (Gradient Boosting Decision Tree): That is, Gradient Boosting Decision Tree, which is an ensemble learning algorithm based on decision trees. It iteratively trains multiple decision trees, and each training optimizes the model based on the residuals of the previous training results, thereby gradually improving the prediction accuracy. It is widely used in regression and classification problems.

[0050] Hierarchical Sampling Strategy: An effective method for processing large-scale datasets. By dividing the dataset into multiple levels and adopting different sampling strategies (such as uniform sampling, weighted sampling, or probability sampling) according to the characteristics of the data at each level, it can reduce the data duplication rate, improve the sampling efficiency, and ensure that the sampling results better meet the actual requirements.

[0051] Apache Flink: An open-source stream processing framework developed by the Apache Software Foundation. Its core is a distributed data stream engine that can execute any stream data program in a data-parallel and pipelined manner, supporting both batch processing and stream processing, with characteristics such as low latency, high throughput, high availability, strict exactly-once consistency guarantee, and powerful state management.

[0052] Robustness: It refers to the ability of a system to maintain its performance and function stability when facing internal structure and external environment changes. That is to say, a robust system can still operate normally and maintain good performance under the influence of various adverse factors such as interference, noise, and faults.

[0053] Sigmoid function: A commonly used logical function whose output value range is restricted between 0 and 1. It is usually used to map any real-number input to this interval to achieve probability output in binary classification problems or as an activation function in neural networks to help the model introduce non-linearity and improve the model's expressive ability.

[0054] Weight factor α: The weight factor α in the prediction function is a key parameter. It is used to adjust the weights of different parts in the function, thereby affecting the prediction result. The magnitude of the α value determines the relative importance of data points or features in the prediction. A larger α value will make the corresponding part have a greater impact on the prediction result, while a smaller α value will reduce its impact. By reasonably setting the α value, the accuracy and robustness of the prediction model can be improved.

[0055] Throughput: An indicator that measures the amount of data successfully processed or transmitted by a system or device per unit time. It reflects the efficiency and ability of the system to process data and is often used to evaluate the performance of computer networks, storage devices, databases, and various information processing systems. High throughput means that the system can complete tasks faster and process more data, which is one of the key indicators for measuring the performance of a system.

[0056] As Figure 1 shown, the embodiment of the present invention provides a fault prediction method, and the steps included are as follows:

[0057] S100: Obtain the real-time data of the system, and determine the first-priority dataset and the second-priority dataset based on the real-time data of the system.

[0058] Obtain the real-time data of the system, and determine the first-priority data set and the second-priority data set based on the hierarchical sampling strategy and data processing; wherein, in the embodiments of the present invention, the first-priority data is high-priority data, including the collected voltage, current, frequency, etc.; the second-priority data is low-priority data, including temperature, power factor, etc.

[0059] Specifically, the grading criteria of the hierarchical sampling strategy are based on the contribution rate of the data to the fault detection and the system resource load.

[0060] Specifically, for high-priority data, a higher sampling frequency is used. Especially when the fault detection is abnormal, the fault detection method in this embodiment will cause the system to switch to the high-frequency sampling mode to ensure the timely tracking of the key data status; for low-priority data, a low-frequency sampling mode is used to reduce the system pressure.

[0061] Specifically, the process of determining the first-priority data set and the second-priority data set based on the real-time data of the system in step S100 can be implemented by the following method:

[0062] S110: Classify the real-time data of the system based on the importance of the data to obtain the first-priority data and the second-priority data.

[0063] Classify the real-time data of the system based on the contribution rate of the data to the fault detection and the system resource load to obtain high-priority data and low-priority data.

[0064] S120: Extract features of the first-priority data by using a first preset method to obtain the first-priority data features.

[0065] Extract synchronous features of the first-priority data to obtain the first-priority data features.

[0066] Specifically, by extracting data features such as the extreme values of voltage and current and the frequency change from the high-priority data, the data features are preferentially processed to generate the first-priority data features for rapid diagnosis.

[0067] In some embodiments, the process of extracting features for high-priority data (such as voltage and current) can be implemented by the following method:

[0068] 1) Use the sliding window technique to segment the real-time data stream. For example, the voltage and current data can use a 1-second window or form a window for every 100 received data.

[0069] 2) Calculate the data within each window through a sliding window, calculating the maximum value, minimum value, and their difference of the current and voltage data, and calculating the frequency change rate (e.g., the frequency difference between adjacent time points). The formula for the instantaneous frequency change rate is: Δf = f(t) - f(t - 1), and the frequency fluctuation amplitude within the window is: Δf_max = max(Δf).

[0070] 3) Further aggregate the features calculated for multiple windows in the second step and set predefined thresholds. When the aggregated features exceed these thresholds, generate first-priority data features.

[0071] S130: Extract features from the second-priority data using a second preset method to obtain second-priority data features.

[0072] Extract the second-priority data in a batch or asynchronous manner to obtain second-priority data features.

[0073] Specifically, extract low-priority features from the low-priority data, including the ambient temperature change rate and the load change rate, using a batch or asynchronous method. After the data is extracted, it is stored in an intermediate buffer and then processed when the system load is low to avoid the high load caused by real-time processing.

[0074] Specifically, batch processing refers to extracting and processing low-priority data regularly (e.g., every hour, every day), which is suitable for features that require long-term trend analysis.

[0075] Specifically, asynchronous processing means that when receiving low-priority data, it is not processed immediately, but stored in a buffer and feature extraction is performed when the system is idle. High-priority data usually does not use batch or asynchronous methods. High-priority data directly affects the stability and security of the system and requires immediate identification and response to anomalies.

[0076] S140: Mark the first-priority data features using a third preset method to construct a first-priority data set.

[0077] Perform real-time marking based on the first-priority data features to construct the first-priority data set D high = {xi}. Real-time marking of high-priority data features is used to train a fast fault detection model. Among them, xi refers to high-priority data points, including: voltage features (maximum voltage value, minimum voltage value, voltage range, etc.), current features (maximum current value, minimum current value, current range, frequency features, etc.), frequency change rate (frequency fluctuation amplitude, etc.).

[0078] S150: Mark the second-priority data features using a fourth preset method to construct a second-priority data set.

[0079] Perform batch marking based on the second-priority data features to construct the second-priority dataset D low ={xi}.

[0080] Wherein, xi refers to low-priority data points, including temperature and power factor.

[0081] The low-priority data is batch-marked during the low load of the system, and is used to train the low-priority dataset model and cooperate with the training of other fault detection models.

[0082] S200: Input the first-priority dataset into the trained fast fault detection model for calculation to obtain the first fault prediction probability.

[0083] Specifically, the fast fault detection model F high ( x) can be trained by the following method:

[0084] Obtain the first sample dataset D′ high ={(xi, yi)}; The first sample dataset includes the first-priority data sample set and the first fault detection result sample;

[0085] Input the first-priority data sample set into the pre-fast fault detection model for calculation to obtain the first sample training result;

[0086] Based on the first fault detection result sample and the first sample training result, update the pre-fast fault detection model, and obtain the fast fault detection model F according to the update result high(x) .

[0087] Specifically, in the first sample dataset D′ high ={(xi, yi)}, xi refers to high-priority data points, including: voltage characteristics, current characteristics, frequency change rate; yi refers to the system state (normal / abnormal).

[0088] Specifically, the first fault prediction probability is calculated by the following formula:

[0089] p high (x)=σ(F high (x))

[0090] Wherein, p high (x) is the result of the fast fault detection model, σ is the sigmoid function, F high(x) is the fast fault detection model, and x is a new data point in the high-priority dataset.

[0091] S300: Determine the relationship between the first fault prediction probability and a preset threshold, and determine whether it is necessary to perform fault prediction based on the second-priority data set according to the determination result.

[0092] If the first fault prediction probability p high (x) is greater than the preset threshold, determine that the fault detection result of the system is a fault. If the first p high (x) is less than the preset threshold: Input the first-priority data set and the second-priority data set into the trained comprehensive fault detection model for calculation, and determine the fault detection result of the system according to the calculation result.

[0093] Specifically, the comprehensive fault detection model F combined(x) can be trained by the following method:

[0094] Obtain a comprehensive sample data set; the comprehensive sample data set includes a first-priority data sample set, a first fault detection result sample, a second-priority data sample set, and a second fault detection result sample;

[0095] Input the first-priority data sample set and the second-priority data sample set into a pre-comprehensive fault detection model for calculation to obtain a comprehensive sample training result;

[0096] Based on the first fault detection result sample, the second fault detection result sample, and the comprehensive sample training result, update the pre-comprehensive fault detection model, and obtain a comprehensive fault prediction model according to the update result.

[0097] Specifically, the comprehensive fault detection model is trained based on the fusion of Dlow and Dhigh data sets, which can improve the prediction accuracy. Moreover, when the system is under low load, it is updated through a batch of low-level priority data to ensure the overall accuracy of the model.

[0098] Specifically, input the first-priority data set and the second-priority data set into the trained comprehensive fault detection model for calculation, and determine the fault detection result of the system according to the calculation result. It can be calculated by the following formula:

[0099] pcombined(x) = α·p high(x) +(1 - α)·p low(x)

[0100] where p combined(x) is the prediction result of the comprehensive model, α is a weight factor for controlling high-priority and low-priority predictions, and p low(x) is the prediction result of inputting the low-priority data set into the low-priority data set model, and p high (x) is the result of inputting the high-priority data set into the fast fault detection model.

[0101] Using the comprehensive model F combined(x) For secondary confirmation, the detection accuracy can be improved and the false alarm rate can be reduced.

[0102] Specifically, the weight factor α is different in the prediction of high-priority data and low-priority data: voltage, current, frequency, etc. are crucial for the real-time monitoring and fault detection of the system. Therefore, higher weights are given to high-priority data; environmental temperature, load change rate, etc. have a smaller direct impact on real-time monitoring compared to high-priority data. Therefore, the weights of low-priority data are usually lower.

[0103] Specifically, the low-priority data set model F low(x) can be trained by the following method:

[0104] Obtain the second sample data set D l ′ ow ={(xi, yi)}; the second sample data set includes the second-priority data sample set and the second fault detection result sample;

[0105] Input the second-priority data sample set into the pre-fast fault detection model for calculation to obtain the second sample training result;

[0106] Based on the second fault detection result sample and the second sample training result, update the pre-fast fault detection model, and obtain the fast fault detection model F according to the update result low(x) .

[0107] Among them, in the second sample data set D l ′ ow ={(xi, yi)}, xi refers to low-priority data points, including temperature and power factor; yi refers to the system state (normal / abnormal).

[0108] Furthermore, the embodiment of the present invention also provides an optimization method strategy for the fast fault detection model, specifically:

[0109] When the fast fault detection model is iteratively trained: use high-priority data to update the fast fault detection model in real time and give priority to fitting the high-priority data features; use low-priority data to supplement and optimize the fast fault detection model to further reduce the error of the model on non-critical features and enhance the robustness of the fast fault detection model.

[0110] Implementing the embodiment of the present invention includes the following beneficial effects:

[0111] 1) In the embodiments of the present invention, by dividing the feature extraction module into high-priority and low-priority levels, the system can classify and process features according to their importance. High-priority features are extracted quickly and transmitted in real time, while low-priority features are processed asynchronously, effectively reducing the instantaneous load of the system and improving the system throughput rate; by simultaneously and parallelly executing the extraction and processing tasks of high-priority and low-priority features, the single-point bottleneck is avoided, and the data processing capacity of the system is increased.

[0112] 2) When an abnormal trend in the distribution network is detected, the present invention can enable the system to quickly switch to the high-frequency sampling mode, quickly obtain key information, and ensure the fast response of the system when a fault occurs; based on the dataset of high-priority features, a fast fault detection model is trained. This model is lightweight and responds quickly, enabling the system to give an instant warning when a fault appears, reducing the response delay, and improving the fault detection efficiency of the system.

[0113] 3) Based on the high-priority data, a comprehensive fault detection model is trained in combination with low-priority features. This model can analyze by integrating multiple types of information, thereby improving the accuracy of fault detection, reducing the risks of false alarms and missed alarms. When the key nodes or the fault judgment is unclear, the secondary confirmation by the comprehensive fault detection model helps to improve the reliability of the judgment, thus making a more accurate fault response and reducing the unnecessary operation and maintenance costs brought by false alarms.

[0114] 4) After the fast fault detection by the system, the fault is verified by the comprehensive fault detection model, reducing the false alarm probability and ensuring that recovery measures are taken only when a real fault occurs, avoiding unnecessary resource waste and interference to the distribution network caused by false alarms; by analyzing the prediction results of the fast fault detection model in detail, the comprehensive fault detection model can early predict potential faults and take preventive measures in advance to ensure the stability and continuity of the distribution network.

[0115] 5) By processing features with different priorities, it can adapt to different requirements in different scenarios, such as the quick recovery in urban distribution networks or the fine prediction and protection in industrial parks. The detection of different-priority data runs independently and cooperates with each other, and can adapt to various distribution network environments. Through the accurate fast fault prediction and secondary confirmation mechanism, the system can reduce the occurrence of false alarms, avoid unnecessary operation and maintenance and equipment switching, thereby reducing the maintenance cost.

[0116] As Figure 2 shown, the embodiments of the present invention also provide a fault detection system, including:

[0117] A first module, which obtains the real-time data of the system and determines a first-priority dataset and a second-priority dataset based on the real-time data of the system;

[0118] The second module inputs the first-priority data set into the trained fast fault detection model for calculation to obtain the first fault prediction probability;

[0119] The third module determines the relationship between the first fault prediction probability and a preset threshold, and determines whether it is necessary to perform fault prediction based on the second-priority data set according to the determination result.

[0120] It can be seen that the content in the above method embodiments is applicable to the system embodiments of the present invention. The functions specifically implemented by the system embodiments of the present invention are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0121] As Figure 3 shown, an embodiment of the present invention further provides a fault detection device, including:

[0122] At least one processor;

[0123] At least one memory for storing at least one program;

[0124] When the at least one program is executed by the at least one processor, the at least one processor implements the fault detection method steps described in the above method embodiments.

[0125] Among them, the memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. The memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a remote memory remotely disposed relative to the processor, and these remote memories may be connected to the processor through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0126] It can be seen that the content in the above method embodiments is applicable to the device embodiments of the present invention. The functions specifically implemented by the device embodiments of the present invention are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0127] In addition, an embodiment of the present application also discloses a computer program product or a computer program. The computer program product or the computer program is stored in a computer-readable storage medium. The processor of the computer device can read the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the above method.

[0128] An embodiment of the present invention also provides a computer-readable storage medium storing a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to implement the above method. Similarly, the content in the above method embodiments is applicable to this storage medium embodiment. The functions specifically implemented by this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0129] It can be understood that all or some of the steps and systems disclosed in the above methods can be implemented as software, firmware, hardware, and their appropriate combinations. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or can be implemented as hardware, or can be implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory, or other memory technologies, CD-ROM, digital versatile disk (DVD), or other optical disk storage, magnetic cassette, tape, magnetic disk storage, or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, a communication medium typically includes computer-readable instructions, data structures, program modules, or other data in a modulated data signal, such as a carrier wave or other transmission mechanism, and can include any information delivery medium.

[0130] The above is a specific description of the preferred embodiment of the present invention. However, the present invention is not limited to the above embodiment. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present invention. These equivalent deformations or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A fault detection method, characterized in that: include: Acquire real-time data of the system, and determine a first priority data set and a second priority data set based on the real-time data of the system; Inputting the first priority data set into a trained fast fault detection model for calculation to obtain a first fault prediction probability; The relationship between the first fault prediction probability and a preset threshold is determined, and whether fault prediction needs to be performed according to the second priority data set is determined according to the result of the determination.

2. The method according to claim 1, characterized in that The determining of the first priority data set and the second priority data set based on the real-time data of the system comprises: Classify the real-time data of the system based on the importance of the data to obtain the first priority data and the second priority data; Extracting features of the first priority data using a first preset method to obtain first priority data features; Extracting features of the second priority data using a second preset method to obtain features of the second priority data; Marking the first priority data features in a third preset manner to construct the first priority data set; The second priority data features are marked using a fourth preset method to construct the second priority data set.

3. The method according to claim 2, characterized in that said extracting features of the first priority data in a first preset manner to obtain first priority data features; The extracting features of the second priority data by using a second preset method to obtain features of the second priority data includes: Performing synchronous feature extraction on the first priority data to obtain features of the first priority data; The second priority data is extracted in batches or asynchronously to obtain the second priority data characteristics.

4. The method according to claim 2, characterized in that The step of using a third preset method to perform feature marking on the first priority data feature to construct the first priority data set; and using a fourth preset method to perform feature marking on the second priority data feature to construct the second priority data set, comprises: Marking the first priority data features in real time to construct a first priority data set; The second priority data features are batch labeled to construct a second priority data set.

5. The method according to claim 1, characterized in that The fast fault detection model is trained by the following method: Acquire a first sample data set; the first sample data set includes a first priority data sample set and a first fault detection result sample; Inputting the first priority data sample set into a pre-fast fault detection model for calculation to obtain a first sample training result; Based on the first fault detection result sample and the first sample training result, the pre-fast fault detection model is updated, and a fast fault detection model is obtained according to the update result.

6. The method according to claim 1, characterized in that The determining, according to the result of the judgment, whether it is necessary to perform fault prediction according to the second priority data set includes: If the result of the judgment is that the first fault prediction probability is greater than the preset threshold, determining that the fault detection result of the system is a fault; If the judgment result is that the first fault prediction probability is less than the preset threshold, the first priority data set and the second priority data set are input into the trained comprehensive fault detection model for calculation, and the fault detection result of the system is determined according to the result of the calculation.

7. The method according to claim 6, characterized in that The comprehensive fault detection model is trained by the following method: Acquire a comprehensive sample data set; the comprehensive sample data set includes a first priority data sample set, a first fault detection result sample, a second priority data sample set, and a second fault detection result sample; Inputting the first priority data sample set and the second priority data sample set into a pre-integrated fault detection model for calculation to obtain a comprehensive sample training result; Based on the first fault detection result sample, the second fault detection result sample and the comprehensive sample training result, the pre-comprehensive fault detection model is updated, and a comprehensive fault prediction model is obtained according to the updated result.

8. A fault detection system, characterized in that: include: A first module, acquiring real-time data of a system, and determining a first priority data set and a second priority data set based on the real-time data of the system; The second module inputs the first priority data set into the trained fast fault detection model for calculation to obtain a first fault prediction probability; The third module determines the relationship between the first fault prediction probability and a preset threshold, and determines whether it is necessary to perform fault prediction based on the second priority data set according to the result of the determination.

9. A fault detection device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to perform the method according to any one of claims 1 to 7 when executed by the processor.

Images