An AI algorithm-based device fault early warning method and system, and a storage medium

Abstract

The application belongs to the technical field of data processing, and particularly relates to a device fault early warning method and system based on an AI algorithm and a storage medium. The method comprises collecting multi-source heterogeneous data to be collected, extracting normal data in a normal state and abnormal data when an anomaly occurs from the multi-source heterogeneous data, generating a first model based on the normal data through machine learning, performing transition learning on the first model to generate a second model, inputting real-time data into the first model to obtain a corresponding prediction value, calculating a deviation degree between the prediction value and the real-time data, judging that the real-time data is abnormal data when the deviation degree is greater than a first threshold value, inputting the real-time data into the second model to obtain an abnormal type, obtaining standard abnormal data corresponding to the abnormal type, calculating a similarity between the standard abnormal data and the real-time data, and judging whether there is a sign of occurrence of the corresponding abnormal type based on the similarity. The application can improve the accuracy of fault identification.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and specifically relates to a device fault early warning method, system and storage medium based on AI algorithm. Background Technology

[0002] With the rapid development of urban rail transit, traditional operation and maintenance models can no longer meet the ever-increasing operational demands. Traditional fault diagnosis and early warning mainly rely on manual inspection and experience-based judgment, which suffers from low efficiency and insufficient accuracy. Existing technologies use artificial intelligence to provide fault warnings for equipment operation status, predicting fault occurrences in advance, reducing system failure rates, and improving system safety and reliability. A similar prior art patent application, Chinese patent application CN117591970A, provides a fault early warning method and system for fully automated assembly equipment. The method includes: real-time acquisition of real-time operating data of the fully automated assembly equipment based on a sensor group; the real-time operating data includes multiple equipment operating parameters; determining the equipment operating data sequence of the fully automated assembly equipment based on the real-time operating data and multiple historical equipment operating data; inputting the equipment operating data sequence into a fault prediction model to determine the fault prediction results of the fully automated assembly equipment for each preset equipment fault type; the fault prediction model includes an LSTM model module; and executing fault early warning operations based on each fault prediction result. However, this method does not consider the possibility of insufficient abnormal training data leading to low model accuracy during model training. A similar prior art is Chinese patent application CN114065510A, which discloses a method, device, and electronic equipment for fault early warning of cooling tower fans. This method acquires historical operating data of the cooling tower fan at different measurement points under healthy operating conditions; based on the historical operating data, it trains a base regressor using a gradient boosting algorithm to obtain a prediction model; it uses the prediction model to predict the baseline values ​​of the target operating parameters of the cooling tower fan; and it provides fault early warning for the cooling tower fan based on the baseline values ​​and actual values ​​of the target operating parameters. However, this method also does not consider the possibility of insufficient abnormal training data leading to low model accuracy during prediction model training. Therefore, this invention provides a method, system, and storage medium for equipment fault early warning based on AI algorithms. Summary of the Invention

[0003] To perform preventative maintenance on rail transit equipment, sufficient data is needed to train the model. Before training the model, the collected data needs to be separated into normal and abnormal data. Usually, the amount of collected data is very large, and extracting normal and abnormal data takes a lot of time. Therefore, it is necessary to extract normal and abnormal data from the collected data quickly and accurately. Secondly, when training the second model, namely the anomaly recognition model, since anomalies occur relatively infrequently in real life, the amount of abnormal data may be insufficient. If the amount of abnormal data is insufficient, the recognition accuracy of the trained model may not be accurate enough. Therefore, it is necessary to generate a sufficient amount of abnormal data to train the anomaly recognition model.

[0004] To achieve the aforementioned objectives, this invention provides a device fault early warning method based on an AI algorithm, which is mainly implemented by performing the following steps:

[0005] Collect multi-source heterogeneous data that needs to be collected. The multi-source heterogeneous data includes multiple feature values, including basic data of the target device, overall status data, component status data, operating data, and log data of the software controlling the target device. Extract normal data and abnormal data when abnormalities occur from the multi-source heterogeneous data. Also, collect the multi-source heterogeneous data that needs to be collected in real time based on a predetermined collection period. The multi-source heterogeneous data collected in real time is called real-time data.

[0006] Based on the normal data, a first model is generated through machine learning. The first model can predict the correlation between data in the normal data under normal conditions. The abnormal data and the predefined abnormal types corresponding to the abnormal data are used as learning data. Transitional learning is performed on the first model to generate a second model. The second model can identify the abnormal types.

[0007] The real-time data is input into the first model for prediction to obtain the corresponding predicted value. The deviation between the predicted value and the real-time data is calculated. When the deviation is less than a preset first threshold, the real-time data is determined to be normal data. When the deviation is greater than the preset first threshold, the real-time data is determined to be abnormal data. The real-time data is then input into the second model to obtain the abnormality type.

[0008] Obtain standard anomaly data corresponding to the anomaly type, calculate the similarity between the standard anomaly data and the real-time data, and if the similarity is greater than a preset second threshold, determine that the real-time data has signs of the corresponding anomaly type, and issue an early warning based on the anomaly type. If the similarity is less than or equal to the second threshold, define a new anomaly type, issue an early warning based on the new anomaly type, and add the new anomaly type and the corresponding real-time data to the learning data for training the second model to retrain and update the second model.

[0009] As a preferred technical solution of the present invention, extracting normal data under normal conditions and abnormal data when anomalies occur from the multi-source heterogeneous data includes the following steps:

[0010] All collected multi-source heterogeneous data are preprocessed, and the processed multi-source heterogeneous data is referred to as first data. According to the acquisition time of the first data, the first data is divided into multiple second data according to a predetermined time interval. Each second data contains multiple first data. The mean and standard deviation of each feature value in the second data are calculated. Based on the mean and standard deviation, each feature value is standardized to obtain a standardized value of each feature value. It is determined whether the standardized value meets a predetermined condition. If the standardized value meets the predetermined condition, a second determination is performed. If the standardized value does not meet the predetermined condition, the first data corresponding to the standardized value is determined to be normal data.

[0011] As a preferred embodiment of the present invention, the predetermined condition is a first condition or a second condition. The first condition is that at least one of the feature values ​​in the first data corresponds to a standardized value that is greater than three times the standard deviation. The second condition is that the number of standardized values ​​in the first data that are greater than two times the standard deviation is greater than a predetermined number. If the standardized value satisfies either the first condition or the second condition, it is determined that the standardized value satisfies the predetermined condition.

[0012] As a preferred embodiment of the present invention, the second determination includes the following steps:

[0013] Calculate the correlation coefficient of each feature value, combine the correlation coefficients of all feature values ​​in the second data into a correlation coefficient matrix, and then obtain the inverse matrix of the correlation coefficient matrix. Based on the inverse matrix and the standardized value, calculate the abnormal value corresponding to the first data. If the abnormal value is greater than a preset third threshold, the first data is determined to be abnormal data; otherwise, the first data is determined to be normal data.

[0014] As a preferred embodiment of the present invention, the calculation of outlier values ​​corresponding to the first data based on the inverse matrix and the standardized values ​​includes the following steps:

[0015] The abnormal value Ab is calculated based on a first formula, which is:

[0016] ,

[0017] in, CM represents the standardized values ​​corresponding to the n feature values ​​in the first data, and CM is the inverse matrix.

[0018] As a preferred embodiment of the present invention, performing transition learning to generate a second model includes the following steps:

[0019] Based on the learning data, transition learning is performed on the first model to generate the second model. Part of the abnormal data is extracted and used as test data to input into the second model. The detection performance of the second model is evaluated and an evaluation value is obtained. If the evaluation value is less than a preset third threshold, a simulation model of the target device is created. Based on the simulation model, various abnormal scenarios are set to cause the simulation model to malfunction, and experimental data under abnormal conditions is generated. The experimental data is collected, and simulation data of the abnormal situation in real conditions is generated based on the experimental data. The simulation data is added to the learning data. This step is repeated to retrain the second model to improve the detection performance of the second model until the evaluation value is greater than or equal to the third threshold.

[0020] As a preferred embodiment of the present invention, calculating the deviation between the predicted value and the real-time data includes the following steps:

[0021] The deviation degree DL is calculated using a second formula, which is:

[0022] ,

[0023] Where n is the total number of feature values ​​in the real-time data. Let i be the i-th actual value in the real-time data. This is the predicted value corresponding to the i-th actual value.

[0024] As a preferred embodiment of the present invention, the standardization of each feature value based on the average value and the standard deviation includes the following steps:

[0025] The standardized value S of each feature value is obtained using the third formula, which is:

[0026] ,

[0027] Where X is a certain characteristic value, AV is the average value of the characteristic value, and D is the standard deviation.

[0028] The present invention also provides an AI-based equipment fault early warning system, comprising the following modules:

[0029] The data collection unit is used to collect multi-source heterogeneous data that needs to be collected. The multi-source heterogeneous data includes multiple feature values, including basic data of the target device, overall status data, component status data, operating data, and log data of the software controlling the target device. Normal data under normal conditions and abnormal data when an anomaly occurs are extracted from the multi-source heterogeneous data. The unit also collects the multi-source heterogeneous data that needs to be collected in real time based on a predetermined collection cycle. The multi-source heterogeneous data collected in real time is called real-time data.

[0030] The model generation unit is used to generate a first model based on the normal data through machine learning. The first model can predict the correlation between data in the normal data under normal conditions. The abnormal data and the predefined abnormal types corresponding to the abnormal data are used as learning data. Transition learning is performed on the first model to generate a second model. The second model can identify the abnormal types.

[0031] An anomaly detection unit is used to input the real-time data into the first model for prediction to obtain the corresponding predicted value, calculate the degree of deviation between the predicted value and the real-time data, determine that the real-time data is normal data when the degree of deviation is less than a preset first threshold, determine that the real-time data is abnormal data when the degree of deviation is greater than the preset first threshold, and input the real-time data into the second model to obtain the anomaly type.

[0032] The type determination unit is used to obtain standard abnormal data corresponding to the abnormal type, calculate the similarity between the standard abnormal data and the real-time data, determine whether the similarity is greater than or equal to a preset second threshold, and if the similarity is greater than the second threshold, determine that the real-time data has signs of the corresponding abnormal type, and issue an early warning based on the abnormal type.

[0033] The present invention also provides a storage medium storing program instructions, wherein the program instructions, when executed, control the device where the storage medium is located to execute any one of the above-described AI algorithm-based device fault early warning methods.

[0034] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0035] This invention extracts normal and abnormal data from multi-source heterogeneous data by performing two precise calculations to extract these data, reducing computational load while maintaining extraction accuracy. A first model is then generated based on the normal data, which generates predicted values ​​for real-time data under normal conditions. A second model is generated based on the abnormal data. During the generation of the second model, considering that insufficient abnormal data can affect its accuracy, more simulated abnormal data is generated to improve the second model's recognition performance and accuracy. Based on both the first and second models, the invention accurately determines whether the target device exhibits signs of abnormality, helping relevant personnel take precise preventative and maintenance measures. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the steps of an AI-based device fault early warning method according to the present invention.

[0037] Figure 2 This is a structural diagram of an AI-based equipment fault early warning system according to the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0039] It is understood that the terms "first," "second," etc., used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are used only to distinguish one element from another. For example, without departing from the scope of this application, a first script may be referred to as a second script, and similarly, a second script may be referred to as a first script.

[0040] This invention provides, for example Figure 1 The device fault early warning method based on AI algorithm shown is mainly implemented by performing the following steps:

[0041] Step S1: Collect the multi-source heterogeneous data that needs to be collected. The multi-source heterogeneous data includes multiple feature values, including the basic data of the target device, the overall status data, the component status data, the operating data, and the log data of the software controlling the target device. Extract the normal data under the normal operating state of the target device and the abnormal data when an anomaly occurs from the multi-source heterogeneous data. Also, collect the multi-source heterogeneous data that needs to be collected in real time based on a predetermined collection cycle. The multi-source heterogeneous data collected in real time is called real-time data.

[0042] Specifically, in order to provide early warning of equipment failures, it is necessary to collect various data related to the target equipment. These data are collectively referred to as multi-source heterogeneous data. Multi-source heterogeneous data includes multiple characteristic values. Taking platform screen doors as an example, basic data includes the opening and closing speeds, pressure resistance ratings, and sealing performance of the platform screen doors. Overall system status data includes the operating mode, operating voltage and current, temperature, operating time, and performance indicators of the platform screen doors. Component status data includes the operating parameters, performance indicators, wear conditions, fault history, electrical parameters, vibration data, sound data, temperature data, and replacement information of various components or parts related to the subway platform screen doors. The data collected includes operational data such as the number of times the shielded doors open and close, opening and closing times, operating speed, and power consumption. The software logs controlling the target equipment include system status, anomalies, and error messages. After acquiring this multi-source heterogeneous data, to train the prediction model, it is necessary to extract normal data under normal conditions and abnormal data under abnormal conditions. A precise and rapid extraction method is used to provide more accurate training data for the subsequent model training. The specific extraction method will be explained in detail later. Real-time data is also collected according to a predetermined collection cycle and input into the subsequently trained model to predict whether there are signs of equipment failure.

[0043] Step S2: Generate a first model based on normal data through machine learning. The first model can predict the correlation between data in normal data under normal conditions. Use abnormal data and the abnormal types corresponding to the predefined abnormal data as learning data to perform transition learning on the first model to generate a second model. The second model can identify the abnormal types.

[0044] Specifically, to obtain a more accurate prediction model, a first model is first generated based on normal data through machine learning. This first model can analyze the correlation between various data when the device is operating normally. In other words, the first model can output the predicted values ​​of various data under normal operating conditions based on the input data. Then, based on the normal model, a second model is generated by overlearning using extracted abnormal data. When training the second model, corresponding abnormal types are defined based on the abnormal data. Training is performed based on the abnormal data and abnormal types to enable the second model to identify abnormal types. If there is not enough abnormal data, more abnormal data is generated to train the second model in order to improve its recognition accuracy, thereby generating a second model with higher recognition performance. The specific process of generating the second model will be explained in detail later.

[0045] Step S3: Input the real-time data into the first model to make a prediction and obtain the corresponding predicted value. Calculate the degree of deviation between the predicted value and the real-time data. If the degree of deviation is less than the preset first threshold, the real-time data is judged to be normal data. If the degree of deviation is greater than the preset first threshold, the real-time data is judged to be abnormal data. Input the real-time data into the second model to obtain the abnormality type.

[0046] Specifically, to determine whether there are any signs of an anomaly, the collected real-time data is first input into the first model to obtain the predicted value corresponding to the real-time data. Then, the deviation between the predicted value and the real-time data is calculated. The specific formula for calculating the deviation is calculated using the subsequent second formula. If the deviation is less than a preset first threshold, it means that the real-time data and the data that should be generated when the equipment is operating normally are not significantly different. Therefore, it is determined that there are no signs of equipment failure at this time. If the deviation is greater than the first threshold, it means that the real-time data and the data that should be generated when the equipment is operating normally are significantly different. Therefore, it is determined that the equipment may have signs of failure at this time. In order to further determine the signs of failure, the real-time data is input into the pre-trained second model for further judgment and to obtain the corresponding anomaly type.

[0047] Step S4: Obtain standard abnormal data corresponding to the abnormal type, calculate the similarity between the standard abnormal data and the real-time data, and if the similarity is greater than the preset second threshold, determine that the real-time data has signs of the corresponding abnormal type, and issue an early warning based on the abnormal type. If the similarity is less than or equal to the second threshold, define a new abnormal type, issue an early warning based on the new abnormal type, and add the new abnormal type and the corresponding real-time data to the learning data for training the second model to retrain and update the second model.

[0048] Specifically, after obtaining the anomaly type, the corresponding standard anomaly data is also obtained. Standard anomaly data refers to the data generated by the target device when the corresponding anomaly type occurred in the past. The similarity between the standard anomaly data and the real-time data is calculated. If the similarity is greater than a preset second threshold, it indicates that the current real-time data and the standard anomaly data are highly similar, which further indicates that the probability of the anomaly situation corresponding to the anomaly type is greater. Based on the anomaly type, an early warning is issued to remind relevant personnel that the above-mentioned anomaly situation may be about to occur. Relevant personnel can take precise preventive and maintenance measures based on the anomaly type indicated. If the similarity is less than or equal to the second threshold, it indicates that the anomaly type predicted by the second model may be a new anomaly type that has not occurred before. Therefore, a new anomaly type is defined, and the new anomaly type and the corresponding real-time data are added to the learning data for training the second model to update the second model. Afterwards, an early warning is also issued based on the new anomaly type to remind relevant personnel to take corresponding preventive measures.

[0049] This invention extracts normal and abnormal data from multi-source heterogeneous data by performing two precise calculations to extract these data, reducing computational load while maintaining extraction accuracy. A first model is then generated based on the normal data, which generates predicted values ​​for real-time data under normal conditions. A second model is generated based on the abnormal data. During the generation of the second model, considering that insufficient abnormal data can affect its accuracy, more simulated abnormal data is generated to improve the second model's recognition performance and accuracy. Based on both the first and second models, the invention accurately determines whether the target device exhibits signs of abnormality, helping relevant personnel take precise preventative and maintenance measures.

[0050] Furthermore, extracting normal data under normal conditions and anomalous data when anomalies occur from multi-source heterogeneous data includes the following steps:

[0051] All collected multi-source heterogeneous data are preprocessed, and the processed multi-source heterogeneous data is referred to as the first data. According to the acquisition time of the first data, the first data is divided into multiple second data according to a predetermined time interval. Each second data contains multiple first data. The mean and standard deviation of each feature value in the second data are calculated. Based on the mean and standard deviation, each feature value is standardized to obtain the standardized value of each feature value. It is determined whether the standardized value meets the predetermined conditions. If the standardized value meets the predetermined conditions, a second judgment is made. If the standardized value does not meet the predetermined conditions, the first data corresponding to the standardized value is determined to be normal data.

[0052] Specifically, to extract normal and abnormal data from multi-source heterogeneous data more accurately and quickly, the multi-source heterogeneous data is first preprocessed. Preprocessing involves cleaning and removing noise and outliers from all collected multi-source heterogeneous data. The processed multi-source heterogeneous data is called the first data. Then, based on the acquisition time of the first data and a predetermined time interval, the first data is divided into multiple second data. For example, if the first data is acquired every 10 minutes, and data from the past 1000 minutes is collected, then there are 100 first data entries. Taking a predetermined time interval of 100 minutes as an example, the first data is divided into 10 parts, and each part of the second data includes 10 first data entries. The second data is A = {A1, A2, A3, ..., A10}, where A1, A2, ..., A10 are all first data entries. Each first data entry also includes multiple... Given a set of eigenvalues ​​A1 = {a1, a2, ..., an}, based on the second data, calculate the mean AV and standard deviation D for each eigenvalue in the second data. Then, standardize each eigenvalue based on the mean and standard deviation to obtain a standardized value. The standardization method will be explained in detail later. Next, determine whether the standardized value meets a predetermined condition. If the predetermined condition is met, it indicates that the first data may be abnormal data. To more accurately determine whether the first data is normal data, a second judgment is performed. The specific method for the second judgment will be explained in detail later. If the standardized value does not meet the predetermined condition, it indicates that the first data is normal data. The first data judged as normal data does not need to undergo a second judgment. The above method reduces the amount of calculation and speeds up the judgment process.

[0053] Furthermore, the predetermined condition is either a first condition or a second condition. The first condition is that there is at least one feature value in the first data whose standardized value is greater than three times the standard deviation. The second condition is that the number of standardized values ​​in the first data that are greater than two times the standard deviation is greater than a predetermined number. If the standardized value satisfies either the first condition or the second condition, it is judged that the standardized value satisfies the predetermined condition.

[0054] Specifically, by using standardized data for judgment, it is possible to quantitatively and easily obtain how many abnormal feature values ​​there are compared with the standard data. If the first condition is met, it means that there are one or more feature values ​​in the first data with a large difference from the standardized value. The presence of data with a large difference indicates that there is a higher probability of anomalies, so the corresponding first data is judged as abnormal data. If the second condition is met, it means that there are a large number of abnormal feature values ​​in the first data. Although the difference is relatively small, the large number of them makes it more likely that there are anomalies. Therefore, if the standard deviation meets either the first or the second condition, it is judged that the standardized value meets the predetermined condition, and the corresponding first data is judged as abnormal data. Otherwise, the corresponding first data is judged as normal data.

[0055] Further, a second judgment is made, including the following steps:

[0056] Calculate the correlation coefficient for each feature value, combine the correlation coefficients of all feature values ​​in the second data into a correlation coefficient matrix, and then obtain the inverse matrix of the correlation coefficient matrix. Based on the inverse matrix and the standardized value, calculate the abnormal value of the corresponding first data. If the abnormal value is greater than the preset third threshold, the first data is judged as abnormal data; otherwise, the first data is judged as normal data.

[0057] Specifically, by filtering out potentially anomalous data in the first judgment, the second judgment can focus on these data without needing to calculate all data, reducing the overall computational load. At the same time, the results of the first calculation are used as a basis for more in-depth analysis and verification, improving the accuracy of the judgment. The elements of the inverse matrix reflect the correlation between the feature values ​​in the original data. When calculating outlier values, the inverse matrix considers the correlation between feature values, thus more accurately calculating the distance between the first data and other first data belonging to the same second data. This distance is used as an indicator to determine whether each first data is an anomalous data. The specific calculation formula for calculating the anomalous value of the corresponding first data based on the inverse matrix and standardized values ​​will be introduced later.

[0058] Furthermore, the outlier values ​​corresponding to the first data are calculated based on the inverse matrix and the standardized values, including the following steps:

[0059] The abnormal value Ab is calculated based on the first formula, which is:

[0060] ,

[0061] in, CM represents the standardized values ​​corresponding to the n eigenvalues ​​in the first dataset, and CM is the inverse matrix.

[0062] Specifically, based on the above formula, by using the inverse of the correlation coefficient matrix, the data is transformed into standardized data, which not only simplifies the computational complexity but also improves computational efficiency.

[0063] Further, transition learning is performed to generate a second model, including the following steps:

[0064] The first model is subjected to transition learning based on the learning data to generate a second model. Some data is extracted from the abnormal data and used as test data to input into the second model. The detection performance of the second model is evaluated and an evaluation value is obtained. If the evaluation value is less than a preset third threshold, a simulation model of the target device is created. Based on the simulation model, various abnormal scenarios are set to make the simulation model malfunction and experimental data under abnormal conditions are generated. The experimental data is collected, and simulation data under abnormal conditions in real situations is generated based on the experimental data. The simulation data is added to the learning data. This step is repeated to retrain the second model to improve its detection performance until the evaluation value is greater than or equal to the third threshold.

[0065] Specifically, to improve the recognition accuracy of the second model, after generating the second model, a portion of abnormal data is selected as test data and input into the second model to evaluate its detection performance and obtain an evaluation value. The evaluation value can be the detection accuracy at a second threshold. If the evaluation value is less than a preset third threshold, it indicates that the recognition accuracy of the currently trained second model is low, possibly due to insufficient abnormal data. Therefore, a simulation model of the target device is created. The simulation model can simulate the working conditions of the target device under real-world conditions. Based on the simulation model, various abnormal scenarios are set to cause the simulation model to malfunction, and experimental data under abnormal conditions is generated. Then, based on the experimental data, simulated data of abnormalities occurring under real-world conditions is generated and added to the learning data for training the second model. This step is repeated to retrain the second model and improve its detection performance until the evaluation value of the second model is greater than or equal to the third threshold. By improving the detection performance and recognition accuracy of the second model through the above method, the second model can more quickly identify the types of anomalies occurring in real-time data, helping relevant personnel to quickly understand the abnormal situation and take relevant preventive measures.

[0066] Furthermore, the degree of deviation between the predicted value and the real-time data is calculated, including the following steps:

[0067] The degree of deviation DL is calculated using the second formula, which is:

[0068] ,

[0069] Where n is the total number of feature values ​​in the real-time data. For the i-th actual value in the real-time data, This is the predicted value corresponding to the i-th actual value.

[0070] Specifically, by calculating the degree of deviation between the predicted value and the real-time data based on the above formula, it is possible to make a preliminary judgment on whether there are potential signs of a failure based on the degree of deviation.

[0071] Furthermore, each feature value is standardized based on its mean and standard deviation, including the following steps:

[0072] The standardized value S of each feature value is obtained using the third formula, which is:

[0073] ,

[0074] Where X is a certain characteristic value, AV is the average value of the characteristic value, and D is the standard deviation.

[0075] According to another aspect of the embodiments of the present invention, reference is made to... Figure 2 As shown, an AI-based equipment fault early warning system is also provided, including a data collection unit, a model generation unit, an anomaly judgment unit, and a type determination unit, to implement the AI-based equipment fault early warning method described above. The specific functions of each module are as follows:

[0076] The data collection unit is used to collect multi-source heterogeneous data that needs to be collected. The multi-source heterogeneous data includes multiple feature values, including basic data of the target device, overall status data, component status data, operating data, and log data of the software controlling the target device. It extracts normal data under normal conditions and abnormal data when anomalies occur from the multi-source heterogeneous data. It also collects the multi-source heterogeneous data that needs to be collected in real time based on a predetermined collection cycle. The multi-source heterogeneous data collected in real time is called real-time data.

[0077] The model generation unit is used to generate a first model based on normal data through machine learning. The first model can predict the correlation between data in normal data under normal conditions. Abnormal data and the anomalous types corresponding to the predefined abnormal data are used as learning data to perform transition learning on the first model to generate a second model. The second model can identify the anomalous types.

[0078] The anomaly detection unit is used to input real-time data into the first model for prediction to obtain the corresponding predicted value, calculate the degree of deviation between the predicted value and the real-time data, and determine that the real-time data is normal data when the degree of deviation is less than a preset first threshold, and that the real-time data is abnormal data when the degree of deviation is greater than the preset first threshold, and input the real-time data into the second model to obtain the anomaly type.

[0079] The type determination unit is used to obtain standard abnormal data corresponding to the abnormal type, calculate the similarity between the standard abnormal data and the real-time data, determine whether the similarity is greater than or equal to a preset second threshold, and if the similarity is greater than the second threshold, determine that the real-time data has signs of the corresponding abnormal type, and issue an early warning based on the abnormal type.

[0080] According to another aspect of the present invention, a storage medium is also provided, which stores program instructions, wherein the program instructions, when executed, control the device where the storage medium is located to execute any of the above-described device fault early warning methods based on AI algorithms.

[0081] In summary, the present invention provides a device fault early warning method, system, and storage medium based on AI algorithms. The method includes collecting multi-source heterogeneous data; extracting normal data under normal conditions and abnormal data when anomalies occur from the multi-source heterogeneous data; generating a first model based on the normal data using machine learning; performing transition learning on the first model to generate a second model; inputting real-time data into the first model for prediction to obtain corresponding predicted values; calculating the deviation between the predicted values ​​and the real-time data; determining that the real-time data is abnormal when the deviation exceeds a first threshold; inputting the real-time data into the second model to obtain the anomaly type; obtaining standard anomaly data corresponding to the anomaly type; calculating the similarity between the standard anomaly data and the real-time data; and determining whether there are signs of the corresponding anomaly type based on the similarity. This invention can improve the accuracy of fault identification.

[0082] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0083] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0084] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0085] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0086] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A device fault early warning method based on AI algorithm, characterized in that, include: Collect multi-source heterogeneous data that needs to be collected. Multi-source heterogeneous data includes multiple feature values, including basic data of the target device, overall status data, component status data, operating data, and log data of the software controlling the target device. Extract normal data under normal operating conditions and abnormal data when anomalies occur from the multi-source heterogeneous data. Also, collect the multi-source heterogeneous data that needs to be collected in real time based on a predetermined collection cycle. The multi-source heterogeneous data collected in real time is called real-time data. The first model is generated based on normal data through machine learning. The first model predicts the correlation between data in normal data under normal conditions. The first model outputs the predicted values ​​of each data under normal operating conditions based on the input data. Abnormal data and the predefined abnormal data corresponding to the abnormal data are used as learning data. Transition learning is performed on the first model to generate a second model. The second model identifies the abnormal types. Real-time data is input into the first model for prediction to obtain the corresponding predicted value. The deviation between the predicted value and the real-time data is calculated. If the deviation is less than a preset first threshold, the real-time data is judged as normal data. If the deviation is greater than the preset first threshold, the real-time data is judged as abnormal data. The real-time data is then input into the second model to obtain the abnormality type. The standard abnormal data corresponding to the abnormality type is obtained. The similarity between the standard abnormal data and the real-time data is calculated. If the similarity is greater than a preset second threshold, the real-time data is judged to have signs of the corresponding abnormality type, and an early warning is issued based on the abnormality type. If the similarity is less than or equal to the second threshold, a new abnormality type is defined, and an early warning is issued based on the new abnormality type. The new abnormality type and the corresponding real-time data are added to the training data of the second model to retrain and update the second model. The process of generating a second model through transitional learning includes: generating a second model by performing transitional learning on the first model based on the learning data; extracting some data from the abnormal data as test data and inputting it into the second model; evaluating the detection performance of the second model and obtaining an evaluation value; if the evaluation value is less than a preset third threshold, creating a simulation model of the target device; setting various abnormal scenarios based on the simulation model to cause the simulation model to malfunction and generating experimental data under abnormal conditions; collecting experimental data; generating simulated data under abnormal conditions in real-world situations based on the experimental data; adding the simulated data to the learning data; repeating this step to retrain the second model and improve its detection performance until the evaluation value is greater than or equal to the third threshold, thereby enabling the second model to quickly identify the type of anomaly occurring in real-time data. When extracting normal and abnormal data from multi-source heterogeneous data, normal and abnormal data are accurately extracted through two calculations, reducing the computational load while ensuring extraction accuracy.

2. The method according to claim 1, characterized in that, Extracting normal data under normal conditions and abnormal data when anomalies occur from multi-source heterogeneous data includes: preprocessing all collected multi-source heterogeneous data, referring to the processed multi-source heterogeneous data as first data; dividing all first data into multiple second data according to the acquisition time of the first data at predetermined time intervals, with each second data containing multiple first data; calculating the mean and standard deviation of each feature value in the second data; standardizing each feature value based on the mean and standard deviation to obtain the standardized value of each feature value; determining whether the standardized value meets predetermined conditions; if the standardized value meets the predetermined conditions, continuing with a second determination; if the standardized value does not meet the predetermined conditions, determining that the first data corresponding to the standardized value is normal data.

3. The method according to claim 2, characterized in that, The predetermined condition is either a first condition or a second condition. The first condition is that there is at least one feature value in the first data that corresponds to a standardized value that is greater than three times the standard deviation. The second condition is that the number of standardized values ​​in the first data that are greater than two times the standard deviation is greater than a predetermined number. If the standardized value satisfies either the first condition or the second condition, it is judged that the standardized value satisfies the predetermined condition.

4. The method according to claim 2, characterized in that, The second judgment includes: calculating the correlation coefficient of each feature value, combining the correlation coefficients of all feature values ​​in the second data into a correlation coefficient matrix, obtaining the inverse matrix of the correlation coefficient matrix, calculating the abnormal value of the corresponding first data based on the inverse matrix and the standardized value, and judging the first data as abnormal data if the abnormal value is greater than the preset third threshold; otherwise, judging the first data as normal data.

5. The method according to claim 4, characterized in that, The calculation of outlier values ​​for the first data point based on the inverse matrix and standardized values ​​includes the following steps: Calculating the outlier value Ab based on the first formula, which is: ,in, CM represents the standardized values ​​corresponding to the n eigenvalues ​​in the first dataset, and CM is the inverse matrix.

6. The method according to claim 1, characterized in that, Calculating the deviation between the predicted value and the real-time data includes: calculating the deviation degree DL using the second formula, which is: Where n is the total number of feature values ​​in the real-time data. For the i-th actual value in the real-time data, This is the predicted value corresponding to the i-th actual value.

7. The method according to claim 2, characterized in that, Each feature value is standardized based on its mean and standard deviation, including: obtaining the standardized value S for each feature value using a third formula, which is: , where X is a certain characteristic value, AV is the average value of the characteristic value, and D is the standard deviation.

8. An AI-based equipment fault early warning system, used to implement the AI-based equipment fault early warning method as described in any one of claims 1-7, characterized in that, include: The data collection unit is used to collect multi-source heterogeneous data that needs to be collected. The multi-source heterogeneous data includes multiple feature values, including basic data of the target device, overall status data, component status data, operating data, and log data of the software controlling the target device. It extracts normal data under normal operating conditions and abnormal data when anomalies occur from the multi-source heterogeneous data. It also collects the multi-source heterogeneous data that needs to be collected in real time based on a predetermined collection cycle. The multi-source heterogeneous data collected in real time is called real-time data. The model generation unit is used to generate a first model based on normal data through machine learning. The first model can predict the correlation between data in normal data under normal conditions. The first model outputs the predicted values ​​of each data under normal operating conditions based on the input data. Abnormal data and the abnormal types corresponding to the predefined abnormal data are used as learning data. Transitional learning is performed on the first model to generate a second model. The second model can identify the abnormal types. The anomaly detection unit is used to input real-time data into the first model for prediction to obtain the corresponding predicted value, calculate the degree of deviation between the predicted value and the real-time data, and determine that the real-time data is normal data when the degree of deviation is less than a preset first threshold, and that the real-time data is abnormal data when the degree of deviation is greater than the preset first threshold, and input the real-time data into the second model to obtain the anomaly type. The type determination unit is used to obtain standard abnormal data corresponding to the abnormal type, calculate the similarity between the standard abnormal data and the real-time data, and determine that the real-time data has signs of the corresponding abnormal type when the similarity is greater than a preset second threshold, and issue an early warning based on the abnormal type. When the similarity is less than or equal to the second threshold, a new abnormal type is defined, an early warning is issued based on the new abnormal type, and the new abnormal type and the corresponding real-time data are added to the training dataset for training the second model to update the second model.

9. A storage medium, characterized in that, The storage medium stores program instructions, wherein, when the program instructions are executed, the device where the storage medium is located is controlled to execute the device fault early warning method based on any one of claims 1-7.

Images