Data processing equipment fault prediction method and system based on Internet of Things

Abstract

The present invention relates to the field of data processing technology, and more specifically to a method and system for predicting faults of data processing equipment based on the Internet of Things. The method comprises: periodically acquiring the operating data of a first detection unit and a second detection unit, and performing cleaning and preprocessing, respectively annotating the first normal data and the first fault data, and training a fault prediction model with the annotated data; creating multiple sample data to train the first prediction model and the second prediction model; respectively acquiring the predicted value of the second detection unit and the predicted value of the first detection unit, comparing the predicted value of the first detection unit with the measured value of the first detection unit, and when the first difference and the second difference are respectively less than their corresponding preset thresholds, determining that the measured values ​​of the first detection unit and the second detection unit are both accurate, annotating them as normal data, inputting the fault prediction model, and obtaining the prediction result. The present invention can improve the accuracy of fault prediction.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more specifically to a method and system for predicting faults of data processing equipment based on the Internet of Things. Background Art

[0002] With the rapid development of IoT technology, more and more devices are connected to the network, forming a huge IoT system. In the IoT system, data processing equipment plays a vital role. However, due to long-term operation of the equipment, environmental factors, and equipment aging, data processing equipment may have various faults. If these faults are not discovered and handled in time, they may lead to serious consequences such as data loss, equipment damage, and even system paralysis.

[0003] Similar prior art includes a Chinese patent application with publication number CN118740598A, which discloses a communication equipment fault prediction and diagnosis method and system based on a large model. The method includes obtaining the operating data of each equipment module in real time based on the transmission channel; according to the preset module arrangement order, the normalized data of each equipment module at each time is counted to obtain a data table; when the fault parameter is detected, the data sub-table before the preset time period is intercepted, and the data sub-table is used as input and the fault parameter is used as output to construct a sample set, and a neural network model is trained based on the sample set; the neural network model is used to assist the detection process of the fault parameter; when the invention is actually used, the trained neural network model is used for fault prediction, which complements the existing fault detection scheme and greatly optimizes the fault handling efficiency. However, the scheme does not consider the issue of whether the sensor measurement value in the device is accurate. Similar prior art also includes a Chinese patent application with publication number CN118644083A, which provides a method and system for predicting equipment failures based on multi-source data analysis, and relates to the field of data analysis technology, including: constructing a loss function through multi-source operation and maintenance data, mining feature concurrent relationships, realizing data-driven modeling, and establishing a periodically updated learning risk prediction model; determining a sensing strategy, configuring a multi-sensor array, using PLC control trigger settings, performing concurrent and fault transmission based on a fiber optic ring network, and performing transmission timestamp constraints based on a network time protocol; generating risk plans based on fault lists, and performing fault management on pre-managed equipment. However, this solution only considers the accuracy of the prediction model, and does not consider whether the sensor measurement values ​​in the equipment are accurate. Therefore, the present invention provides a method and system for predicting failures of data processing equipment based on the Internet of Things. Summary of the invention

[0004] In order to better solve the above problems, the present invention provides a data processing equipment fault prediction method based on the Internet of Things, wherein multiple groups of sensors are installed inside or outside the data processing equipment, each group of sensors includes a first detection unit and a second detection unit, wherein the first detection unit and the second detection unit are sensors of different types but have correlation, based on the shape of the data processing equipment, the first detection unit and the second detection unit are symmetrically installed about the central axis on the data processing equipment, and a preset threshold is set for the difference between the measured value and the predicted value obtained by each detection unit, respectively, the detection unit refers to any one or more of the first detection unit and the second detection unit, and is implemented by executing the following steps:

[0005] Step S1: periodically acquiring operation data of the first detection unit and the second detection unit, and cleaning and preprocessing the operation data to acquire the preprocessed operation data, wherein the operation data is divided into first normal data and first fault data, the first normal data and the first fault data are annotated respectively, a fault prediction model is created, and the annotated first normal data and the first fault data are used as input data, and the fault type and the corresponding fault cause are used as output results of the fault prediction model to train the fault prediction model;

[0006] Step S2: creating a plurality of sample data based on the first normal data, and dividing the sample data into first sample data and second sample data according to corresponding detection units, and creating a first prediction model and a second prediction model based on a neural network, training the first prediction model using the first sample data, training the second prediction model using the second sample data, and acquiring the trained first prediction model and the second prediction model;

[0007] Step S3: obtaining the measured values ​​of the first detection unit and the second detection unit, inputting the measured values ​​of the first detection unit and the second detection unit into the first prediction model and the second prediction model respectively, and obtaining the predicted value of the second detection unit and the predicted value of the first detection unit respectively, comparing the predicted value of the first detection unit with the measured value of the first detection unit to obtain a first difference, and comparing the predicted value of the second detection unit with the measured value of the second detection unit to obtain a second difference;

[0008] Step S4: When the first difference and the second difference are respectively smaller than the corresponding preset thresholds, it is determined that the measured values ​​of the first detection unit and the second detection unit are accurate and annotated as normal data, and the measured values ​​of the first detection unit and the second detection unit are input into the fault prediction model to obtain a prediction result, which includes the fault type and the corresponding fault cause.

[0009] As a preferred technical solution of the present invention, the step S4 further includes:

[0010] Step S5: When the second difference is greater than or equal to the corresponding preset threshold, the measured value of the first detection unit is replaced by the predicted value of the first detection unit, and re-entered into the first prediction model to obtain the second predicted value of the second detection unit; when the difference between the second predicted value of the second detection unit and the measured value of the second detection unit is less than the corresponding preset threshold, it is determined that the measured value of the first detection unit is inaccurate, and the measured value of the first detection unit is replaced by the predicted value of the first detection unit, and then input into the fault prediction model together with the measured value of the second detection unit to obtain a prediction result.

[0011] As a preferred technical solution of the present invention, the step S5 further includes:

[0012] Step S6: When the difference between the second predicted value of the second detection unit and the actual measured value of the second detection unit is greater than or equal to the corresponding preset threshold, the predicted value of the second detection unit is used to replace the actual measured value of the second detection unit, and the predicted value is re-entered into the second prediction model to obtain the second predicted value of the first detection unit. When the difference between the second predicted value of the first detection unit and the actual measured value of the first detection unit is less than the corresponding preset threshold, it is determined that the actual measured value of the second detection unit is inaccurate, and the predicted value of the second detection unit is used to replace the actual measured value of the second detection unit, and then combined with the actual measured value of the first detection unit, it is input into the fault prediction model to obtain a prediction result.

[0013] As a preferred technical solution of the present invention, the step S6 further includes:

[0014] Obtaining the measured values ​​and predicted values ​​of the detection units in other groups, and when the differences between the measured values ​​and predicted values ​​of two or more groups of detection units are greater than or equal to the corresponding preset thresholds, detecting the setting information of the data processing device, otherwise determining that the measured values ​​of the first detection unit and the second detection unit are both inaccurate;

[0015] When any one or both of the measured values ​​of the first detection unit and the second detection unit are inaccurate, the corresponding measured values ​​are annotated as abnormal data and used as second abnormal data to continue training the fault prediction model.

[0016] As a preferred technical solution of the present invention, detecting the setting information of the data processing device includes:

[0017] When the setting information of the data processing device changes, the multiple groups of measured values ​​of the detection units are annotated as normal data and used as the second normal data to retrain the first prediction model and the second prediction model. When the setting information of the data processing device does not change, the multiple groups of measured values ​​of the detection units are annotated as abnormal data and used as the second abnormal data to continue training the fault prediction model.

[0018] As a preferred technical solution of the present invention, in step S2, dividing the sample data into first sample data and second sample data according to the corresponding detection units includes:

[0019] Each of the sample data includes the measured value of the first detection unit and the measured value of the second detection unit. The measured value of the first detection unit in the sample data is used as the first sample data, and the measured value of the second detection unit in the sample data is used as the second sample data. The measured value of the first detection unit and the measured value of the second detection unit correspond to each other at a time point.

[0020] As a preferred technical solution of the present invention, in step S2, obtaining the trained first prediction model and the second prediction model includes:

[0021] Step S21: inputting the first sample data as input data into the first prediction model, and training the first prediction model. During the training process, the first prediction model learns how to map the measured value of the first detection unit to the measured value of the second detection unit at its corresponding time point, and uses the output result of the first prediction model as the predicted value of the second detection unit;

[0022] Step S22: comparing the error between the output result of the first prediction model and the actual measured value of the second detection unit, and adjusting the parameters of the first prediction model based on the error, so that the output result of the first prediction model is equal to the actual measured value of the second detection unit at the same time point corresponding to the actual measured value of the first detection unit in the first sample data;

[0023] Step S23: using the actual measured value of the first detection unit and the actual measured value of the second detection unit under normal operating conditions obtained in real time as verification data, verifying the output result of the first prediction model based on the verification data, and obtaining the first prediction model when the error between the output result of the first prediction model and the actual measured value of the second detection unit is less than or equal to a set threshold;

[0024] Step S24: Repeat the method from step S21 to step S23 to obtain the second prediction model.

[0025] As a preferred technical solution of the present invention, in step S3, obtaining the predicted value of the second detection unit and the predicted value of the first detection unit includes:

[0026] The measured value of the first detection unit is input as input data into the first prediction model, and the first prediction model outputs the predicted value of the second detection unit. The measured value of the second detection unit is input as input data into the second prediction model, and the second prediction model outputs the predicted value of the first detection unit.

[0027] The present invention also provides a data processing equipment fault prediction system based on the Internet of Things as described above, comprising the following modules:

[0028] a data acquisition unit, configured to periodically acquire operation data of the first detection unit and the second detection unit, clean and preprocess the operation data, acquire the preprocessed operation data, divide the operation data into first normal data and first fault data, annotate the first normal data and the first fault data respectively, create a fault prediction model, and use the annotated first normal data and the first fault data as input data, and the fault type and the corresponding fault cause as output results of the fault prediction model, to train the fault prediction model;

[0029] a model training unit, configured to create a plurality of sample data, and divide the sample data into first sample data and second sample data according to corresponding detection units, and further create a first prediction model and a second prediction model based on a neural network, train the first prediction model using the first sample data, train the second prediction model using the second sample data, and obtain the trained first prediction model and the second prediction model;

[0030] a prediction unit, used to obtain the measured values ​​of the first detection unit and the second detection unit, input the measured values ​​of the first detection unit and the second detection unit into a first prediction model and a second prediction model respectively, and obtain the predicted value of the second detection unit and the predicted value of the first detection unit respectively, compare the predicted value of the first detection unit with the measured value of the first detection unit to obtain a first difference, and compare the predicted value of the second detection unit with the measured value of the second detection unit to obtain a second difference;

[0031] A calculation unit is used to determine that the measured values ​​of the first detection unit and the second detection unit are accurate and annotate them as normal data when the first difference and the second difference are respectively less than the corresponding preset thresholds, input the measured values ​​of the first detection unit and the second detection unit into the fault prediction model, and obtain a prediction result, wherein the prediction result includes a fault type and a corresponding fault cause.

[0032] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0033] 1. The technical solution of the present invention periodically obtains the operating data of the first detection unit and the second detection unit, cleans and preprocesses them, and annotates the first normal data and the first fault data respectively, thereby ensuring the quality of the data and providing accurate training data for the fault prediction model. A fault prediction model is created, and the annotated data is used as input data to train the fault prediction model; so that the behavior pattern of future data processing equipment can be identified, providing a basis for timely handling of faults. In addition, by creating multiple sample data, and dividing the sample data into first sample data and second sample data according to the corresponding detection units, a first prediction model and a second prediction model are created based on a neural network, the first sample data is used to train the first prediction model, and the second sample data is used to train the second prediction model, and the trained first prediction model and second prediction model are obtained; so that the measurement values ​​between detection units with correlation can be accurately predicted, providing a basis for fault detection.

[0034] 2. The technical solution of the present invention can timely discover the state of the detection unit under the current conditions and discover the potential failure of the data processing device in advance by obtaining the measured values ​​of the first detection unit and the second detection unit, inputting the measured values ​​of the first detection unit and the second detection unit into the first prediction model and the second prediction model respectively, and obtaining the predicted value of the second detection unit and the predicted value of the first detection unit respectively, comparing the predicted value of the first detection unit with the measured value of the first detection unit to obtain the first difference, and comparing the predicted value of the second detection unit with the measured value of the second detection unit to obtain the second difference. When only the measured value of one of the detection units is obtained, the prediction result of the data processing device can be obtained earlier through mutual prediction. When the first difference and the second difference are respectively less than the corresponding preset thresholds, it is determined that the measured values ​​of the first detection unit and the second detection unit are accurate and annotated as normal data, and the measured values ​​of the first detection unit and the second detection unit are input into the fault prediction model to obtain the prediction result, which includes the fault type and the corresponding fault cause. Through the coordination of the above steps, the present invention can eliminate the problem of inaccurate measurement values ​​caused by damage to the detection unit in the equipment, thereby avoiding the adverse effects of such inaccurate measurement values ​​on fault prediction, thereby improving the accuracy of fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0036] Figure 1 It is a flowchart of the steps of the method for predicting faults of data processing equipment based on the Internet of Things in the present invention;

[0037] Figure 2 This is a structural diagram of the composition of the data processing equipment fault prediction system based on the Internet of Things in the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present 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 only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0040] With the rapid development of IoT technology, more and more devices are connected to the network, forming a huge IoT system. In the IoT system, data processing equipment plays a vital role. However, due to long-term operation of the equipment, environmental factors, and equipment aging, data processing equipment may have various faults. If these faults are not discovered and handled in time, they may lead to serious consequences such as data loss, equipment damage, and even system paralysis.

[0041] In view of the above technical problems, the present invention proposes Figure 1The data processing equipment fault prediction method based on the Internet of Things shown in the figure is to install multiple groups of sensors inside or outside the data processing equipment, each group of sensors includes a first detection unit and a second detection unit, wherein the first detection unit and the second detection unit are sensors of different types but with correlation, based on the shape of the data processing equipment, the first detection unit and the second detection unit are symmetrically installed on the data processing equipment about the central axis, and a preset threshold is set for the difference between the measured value and the predicted value obtained by each detection unit, the detection unit refers to any one or more of the first detection unit and the second detection unit, and is implemented by executing the following steps:

[0042] Step S1: periodically obtain the operation data of the first detection unit and the second detection unit, clean and preprocess the operation data, obtain the preprocessed operation data, divide the operation data into first normal data and first fault data, annotate the first normal data and the first fault data respectively, create a fault prediction model, and use the annotated first normal data and the first fault data as input data, and the fault type and the corresponding fault cause as the output result of the fault prediction model to train the fault prediction model.

[0043] Specifically, in order to predict the fault, it is necessary to first train the fault prediction model, and the accuracy of the fault prediction model prediction depends on the support of high-quality training data. First, the operation data of the first detection unit and the second detection unit are periodically collected, and the operation data includes but is not limited to temperature, fan speed, power supply voltage, current and vibration of the hard disk drive and other characteristics related to the performance of the data processing device. However, the original data obtained from the detection unit may contain noise, missing value data and other problems, so it is necessary to remove noise and outliers, and then perform preprocessing, such as normalization processing to improve data quality. The data preprocessing method belongs to the prior art and will not be repeated here. The preprocessed operation data is also divided into normal data and fault data, and the normal data and fault data are annotated. The fault prediction model is trained using the annotated normal data and fault data, such as using a machine learning algorithm (support vector machine, random forest, etc.). The first normal data and the first abnormal data are the data for training the model first, and the second normal data or the second abnormal data may be used to train the model later. Normal data refers to data generated by the data processing equipment in normal operation that satisfies the correlation between the same group of sensors. Abnormal data refers to data between the measurement values ​​obtained by the first detection unit and the second detection unit that do not satisfy the correlation between the same group of sensors. The correlation means that the measurement values ​​of the first detection unit and the measurement values ​​of the second detection unit can predict each other. The specific method will be described in detail below. Through periodic data collection and preprocessing, the quality of the data is ensured, and accurate training data is provided for the fault prediction model. The trained fault prediction model can identify the behavior pattern of the future data processing equipment and provide a basis for timely handling of faults.

[0044] Step S2: Create multiple sample data based on the first normal data, and divide the sample data into first sample data and second sample data according to the corresponding detection units, and create a first prediction model and a second prediction model based on the neural network, use the first sample data to train the first prediction model, use the second sample data to train the second prediction model, and obtain the trained first prediction model and second prediction model.

[0045] Specifically, multiple sample data are created based on the first normal data, and these sample data are divided into: first sample data such as temperature data and second sample data such as fan speed data. The first sample data and the second sample data are correlated. If the temperature in the data processing device rises, the fan in the cooling system will be triggered to increase the speed to dissipate heat. Therefore, an increase in the reading of the temperature sensor can predict an increase in the reading of the fan speed sensor, and a corresponding increase in the reading of the fan speed sensor can also predict an increase in the reading of the temperature sensor. Two prediction models are created using neural network technology: a first prediction model and a second prediction model. The first sample data is used to train the first prediction model so that it can predict the fan speed data based on the temperature data; the second sample data is used to train the second prediction model so that it can predict the temperature data based on the fan speed data. The trained prediction model can accurately predict the measurement values ​​between detection units with correlation, providing a basis for fault detection.

[0046] Step S3: Obtain the measured values ​​of the first detection unit and the second detection unit, input the measured values ​​of the first detection unit and the second detection unit into the first prediction model and the second prediction model respectively, and obtain the predicted value of the second detection unit and the predicted value of the first detection unit respectively, compare the predicted value of the first detection unit with the measured value of the first detection unit to obtain a first difference, compare the predicted value of the second detection unit with the measured value of the second detection unit to obtain a second difference.

[0047] Specifically, the measured values ​​of the first detection unit and the second detection unit are obtained in real time, such as the measured temperature value obtained by the temperature detection unit and the measured fan speed value obtained by the fan speed detection unit, and the measured temperature value is input into the trained first prediction model to obtain the predicted value of the fan speed, and the measured fan speed value is input into the trained second prediction model to obtain the predicted value of the temperature. By comparing the measured value and the predicted value of the detection unit, it is possible to timely discover the situation in which the detection unit obtains the measured value under the current conditions, such as inaccurate measured data caused by damage to the detection unit, so as to discover the potential faults that may occur in the data processing equipment in advance, so as to identify the factors affecting the fault prediction as early as possible.

[0048] Step S4: When the first difference and the second difference are respectively smaller than their corresponding preset thresholds, it is determined that the measured values ​​of the first detection unit and the second detection unit are accurate, and annotated as normal data, and the measured values ​​of the first detection unit and the second detection unit are input into the fault prediction model to obtain the prediction result, which includes the fault type and the corresponding fault cause.

[0049] Specifically, by comparing the predicted value with the measured value obtained by the detection unit, it is possible to timely discover whether the detection unit is faulty, thereby improving the accuracy of fault prediction. For example, the temperature measured value obtained by the first detection unit ranges from 45 to 65 degrees. According to the correlation between temperature and fan speed, the corresponding fan speed measured value of the second detection unit should be between 34 and 67 rpm under normal operating conditions. If the measured temperature of the first detection unit is 50 degrees at this time, the predicted output result is 65 rpm through the first prediction model prediction, and the measured fan speed of the second detection unit is 60 rpm. The predicted output result is 58 degrees through the second prediction model prediction, which is the predicted value of the first detection unit. At this time, the difference between the predicted value and the measured value of the first detection unit is less than the corresponding preset threshold value (for example, the preset threshold value corresponding to the difference between the measured value and the predicted value of the first detection unit is defined as 8 degrees), indicating that the measured data obtained by the first detection unit and the second detection unit are accurate, and can be directly input into the fault prediction model for fault prediction, obtain the prediction result, and notify the relevant staff to take different countermeasures based on the prediction result.

[0050] Through the coordination of the above steps, the present invention can eliminate the problem of inaccurate measurement values ​​caused by damage to the detection unit in the equipment, thereby avoiding the adverse effects of such inaccurate measurement values ​​on fault prediction, thereby improving the accuracy of fault prediction.

[0051] Furthermore, after the above step S4, the following steps are further included:

[0052] Step S5: When the second difference is greater than or equal to the corresponding preset threshold, the measured value of the first detection unit is replaced by the predicted value of the first detection unit, and re-entered into the first prediction model to obtain the second predicted value of the second detection unit. When the difference between the second predicted value of the second detection unit and the measured value of the second detection unit is less than the corresponding preset threshold, it is determined that the measured value of the first detection unit is inaccurate, and the measured value of the first detection unit is replaced by the predicted value of the first detection unit, and then input into the fault prediction model together with the measured value of the second detection unit to obtain a prediction result.

[0053] Specifically, when the second difference is greater than or equal to the corresponding preset threshold, it is possible that one of the detection units is damaged or faulty. For example, the measured value obtained by the first detection unit is 88 degrees, and the measured value obtained by the second detection unit is 65 rpm. After the measured value of 88 degrees obtained by the first detection unit is input into the first prediction model, the output result is 220 rpm, that is, the second difference is greater than the corresponding preset threshold, but at this time it is still not possible to determine whether the first detection unit is faulty or the second detection unit in the same group is faulty. Through the above method, the measured value of the first detection unit is replaced by the predicted value of the first detection unit output by the second prediction model, and the predicted value of the second detection unit is compared with the measured value of the second detection unit again, and it is determined that the first detection unit is faulty. When the measured value is inaccurate, the accuracy of the fault prediction model prediction will also be reduced. Therefore, by replacing the inaccurate measured value with the predicted value, the data quality input to the fault prediction model is optimized, thereby improving the accuracy of fault prediction.

[0054] The above method is when the measurement values ​​of the first detection unit and the second detection unit are obtained at the same time. The present invention also includes another embodiment. In order to obtain the prediction result earlier, when only the measurement value of either the first detection unit or the second detection unit is obtained, based on the correlation between the first detection unit and the second detection unit, for example, when only the measured value of the first detection unit is obtained, the measured value of the first detection unit is input into the first prediction model, and the first prediction model outputs the predicted value of the second detection unit; then the predicted value of the second detection unit output by the first prediction model is input into the second prediction model as input data, and the second prediction model outputs the predicted value of the first detection unit; at the same time, the predicted value of the second detection unit can be obtained through the measured value of the second detection unit, and then the difference between the measured value and the predicted value is compared. This method can obtain the prediction result of the data processing device earlier when only the measurement value of one detection unit is obtained.

[0055] Furthermore, after the above step S5, the following steps are further included:

[0056] Step S6: When the difference between the second predicted value of the second detection unit and the actual measured value of the second detection unit is greater than or equal to the corresponding preset threshold value, the actual measured value of the second detection unit is replaced by the predicted value of the second detection unit and re-entered into the second prediction model to obtain the second predicted value of the first detection unit. When the difference between the second predicted value of the first detection unit and the actual measured value of the first detection unit is less than the corresponding preset threshold value, it is determined that the actual measured value of the second detection unit is inaccurate, and the actual measured value of the second detection unit is replaced by the predicted value of the second detection unit, and then input into the fault prediction model together with the actual measured value of the first detection unit to obtain the prediction result.

[0057] Specifically, there is another situation. When the difference between the second predicted value of the second detection unit and the actual measured value of the second detection unit is greater than or equal to the corresponding preset threshold, it cannot be determined whether the second detection unit is faulty or both the first detection unit and the second detection unit are faulty. In order to distinguish this situation, the second predicted value of the first detection unit is obtained by the above method, and then compared. If at this time the difference between the second predicted value of the first detection unit and the actual measured value of the first detection unit is less than the corresponding preset threshold, it is judged that the second detection unit is faulty.

[0058] Furthermore, after the above step S6, the following steps are further included:

[0059] Obtaining the measured values ​​and predicted values ​​of the detection units in other groups, and when the differences between the measured values ​​and the predicted values ​​of two or more groups of detection units are greater than or equal to their corresponding preset thresholds, detecting the setting information of the data processing device, otherwise determining that the measured values ​​of the first detection unit and the second detection unit are both inaccurate;

[0060] When any one or both of the measured values ​​of the first detection unit and the second detection unit are inaccurate, the corresponding measured value is annotated as abnormal data and used as the second abnormal data to continue training the fault prediction model.

[0061] Specifically, it is also possible that both detection units fail or more detection units fail. If the measured values ​​of the detection units in other groups are also inaccurate, it is also possible that the correlation between the detection units has changed. To verify this situation, through the above method, when the difference between the measured values ​​and the predicted values ​​of two or more groups of detection units is greater than or equal to their corresponding preset thresholds, it is necessary to detect the setting information of the data processing device to verify whether the correlation between the detection units has changed. If the correlation has changed, further processing is required. The detailed detection method will be described below.

[0062] Furthermore, the setting information of the detection data processing device includes:

[0063] When the setting information of the data processing device changes, the measured values ​​of multiple groups of detection units are annotated as normal data and used as the second normal data to retrain the first prediction model and the second prediction model. When the setting information of the data processing device does not change, the measured values ​​of multiple groups of detection units are annotated as abnormal data and used as the second abnormal data to continue training the fault prediction model.

[0064] Specifically, if the measured values ​​of multiple detection units and their corresponding predicted values ​​are not accurate, it may also be that the operating state of the device has changed, or the setting information of the device has been adjusted, for example, the heat dissipation mode is turned on, which means that the correlation between the detection units has changed, and the existing model is no longer applicable. By performing the detection of the device setting information, check whether the configuration parameters of the data processing device have changed, such as workload, performance parameters, environmental conditions, etc. If the setting information of the device has changed, it means that the change of the measured value obtained by the detection unit may be caused by the change of normal operating conditions, rather than the failure of the detection unit. In this case, these inaccurate measured values ​​are annotated as normal data and used as the second normal data. The first prediction model and the second prediction model are retrained using these newly obtained normal data as the second normal data to update the first prediction model and the second prediction model to adapt to the new correlation. For example, the larger the current, the higher the temperature. Now the heat dissipation mode is enabled, the larger the voltage and current, the greater the wind force, and the lower the temperature. If the setting information of the data processing device has not changed, the inaccuracy of the measured values ​​and their predicted values ​​of multiple detection units may be caused by potential failures. In this case, these inaccurate measured values ​​are annotated as abnormal data and used as the second abnormal data. Retraining the fault prediction model using these newly acquired abnormal data can further improve the fault prediction model's ability to identify new fault modes to better identify new or unknown fault modes. By using the setting information of the detection equipment, it is possible to distinguish between changes in the measurement values ​​of the detection unit caused by changes in normal operating conditions and real fault conditions, thereby improving the accuracy of fault prediction.

[0065] Furthermore, in the above step S2, dividing the sample data into first sample data and second sample data according to the corresponding detection units includes:

[0066] Each sample data includes the measured value of the first detection unit and the measured value of the second detection unit. The measured value of the first detection unit in the sample data is used as the first sample data, and the measured value of the second detection unit in the sample data is used as the second sample data. The measured value of the first detection unit and the measured value of the second detection unit correspond to each other at a time point.

[0067] Specifically, the measured value of the first detection unit in each sample data is extracted to form a first sample data set, which is used to train the first prediction model, the goal of which is to predict the measured value of the second detection unit. Similarly, the measured value of the second detection unit in each sample data is extracted to form a second sample data set, which is used to train the second prediction model, the goal of which is to predict the measured value of the first detection unit. For training an accurate prediction model, it is necessary to ensure that the first sample data and the second sample data correspond to each other at the time point, and the training data directly reflects the behavior of the two sensors under the same conditions, because the prediction model learns the correlation between the two sensors at the same time point. The data partitioning method allows the system to better adapt to the interactions and potential changes between the detection units and improves the system's sensitivity to changes in equipment status. When the actual measurement value is inconsistent with the predicted value, it can be more easily identified which sensor or which prediction model may need to be adjusted or retrained. The sample data partitioning method provides a solid foundation for establishing an accurate prediction model, which helps to ultimately improve the accuracy of fault prediction.

[0068] Further, in step S2, obtaining the trained first prediction model and the second prediction model includes:

[0069] Step S21: inputting the first sample data as input data into the first prediction model, and training the first prediction model. During the training process, the first prediction model learns how to map the measured value of the first detection unit to the measured value of the second detection unit at the corresponding time point, and uses the output result of the first prediction model as the predicted value of the second detection unit;

[0070] Step S22: comparing the error between the output result of the first prediction model and the actual measured value of the second detection unit, and adjusting the parameters of the first prediction model based on the error, so that the output result of the first prediction model is equal to the actual measured value of the second detection unit at the same time point corresponding to the actual measured value of the first detection unit in the first sample data;

[0071] Step S23: using the actual measured value of the first detection unit and the actual measured value of the second detection unit under normal operating conditions obtained in real time as verification data, verifying the output result of the first prediction model based on the verification data, and obtaining the first prediction model when the error between the output result of the first prediction model and the actual measured value of the second detection unit is less than or equal to a set threshold;

[0072] Step S24: Repeat the method from step S21 to step S23 to obtain a second prediction model.

[0073] Specifically, the first sample data (the measured value of the first detection unit) is used as input data and input into the first prediction model. During the training process, the first prediction model learns how to map the measured value of the first detection unit to the measured value of the second detection unit at the corresponding time point through a machine learning algorithm. The output result of the first prediction model will be used as the predicted value of the second detection unit. The error between the output result of the first prediction model (the predicted value of the second detection unit) and the actual measured value of the second detection unit is compared. Based on this error, the parameters of the first prediction model are adjusted, and an optimization algorithm (such as gradient descent) is used to minimize the prediction error, with the goal of making the output result of the first prediction model as close as possible to the measured value of the second detection unit at the same time point corresponding to the measured value of the first detection unit in the first sample data. The measured values ​​of the first detection unit and the second detection unit under normal operating conditions obtained in real time are collected as verification data. These verification data are used to evaluate the output result of the first prediction model. If the error between the output result of the first prediction model and the measured value of the second detection unit is less than or equal to the preset second threshold, it is considered that the model has reached sufficient accuracy and can be obtained and used. Repeat the method of steps S21 to S23, use the second sample data (measured value of the second detection unit) as input data, and train the second prediction model. The second prediction model learns how to map the measured value of the second detection unit to the measured value of the first detection unit at the corresponding time point. By comparing, adjusting parameters and verifying, an accurate second prediction model is finally obtained. Through the training and verification process, it is ensured that the prediction model can accurately predict the measured values ​​of the detection units with related relationships, improve the accuracy of fault detection, and provide a solid foundation for fault prediction.

[0074] Furthermore, in the above step S3, obtaining the predicted value of the second detection unit and the predicted value of the first detection unit includes:

[0075] The measured value of the first detection unit is input as input data into the first prediction model, and the first prediction model outputs the predicted value of the second detection unit. The measured value of the second detection unit is input as input data into the second prediction model, and the second prediction model outputs the predicted value of the first detection unit.

[0076] Specifically, to obtain the mutual prediction value between the first detection unit and the second detection unit, first input the measured value of the first detection unit as input data into the first prediction model. As mentioned above, the task of the first prediction model is to learn how to predict the predicted value of the second detection unit based on the data of the first detection unit. Therefore, the trained first prediction model will output the predicted value of the second detection unit. The measured value of the second detection unit can also be input as input data into the second prediction model. As mentioned above, the task of the second prediction model is to learn how to predict the predicted value of the first detection unit based on the data of the second detection unit. Therefore, the second prediction model will output the predicted value of the first detection unit. At the same time, this method can also obtain the corresponding predicted value of another detection unit through the predicted values ​​of the first detection unit and the second detection unit, so as to facilitate more accurate fault prediction.

[0077] The present invention also provides a Figure 2 The data processing equipment fault prediction system based on the Internet of Things shown includes the following modules:

[0078] a data acquisition unit, used to periodically acquire the operation data of the first detection unit and the second detection unit, clean and preprocess the operation data, acquire the preprocessed operation data, divide the operation data into first normal data and first fault data, annotate the first normal data and the first fault data respectively, create a fault prediction model, and use the annotated first normal data and the first fault data as input data, and the fault type and the corresponding fault cause as output results of the fault prediction model to train the fault prediction model;

[0079] A model training unit, used to create a plurality of sample data, and divide the sample data into first sample data and second sample data according to the corresponding detection units, and also create a first prediction model and a second prediction model based on a neural network, train the first prediction model using the first sample data, train the second prediction model using the second sample data, and obtain the trained first prediction model and the second prediction model;

[0080] A prediction unit, used to obtain the measured values ​​of the first detection unit and the second detection unit, input the measured values ​​of the first detection unit and the second detection unit into the first prediction model and the second prediction model respectively, and obtain the predicted value of the second detection unit and the predicted value of the first detection unit respectively, compare the predicted value of the first detection unit with the measured value of the first detection unit to obtain a first difference, and compare the predicted value of the second detection unit with the measured value of the second detection unit to obtain a second difference;

[0081] A calculation unit is used to determine that the measured values ​​of the first detection unit and the second detection unit are accurate and annotate them as normal data when the first difference and the second difference are respectively less than their corresponding preset thresholds, input the measured values ​​of the first detection unit and the second detection unit into the fault prediction model, and obtain the prediction result, which includes the fault type and the corresponding fault cause.

[0082] In summary, the present invention periodically obtains the operating data of the first detection unit and the second detection unit, cleans and preprocesses them, and annotates the first normal data and the first fault data respectively, thereby ensuring the quality of the data and providing accurate training data for the fault prediction model. A fault prediction model is created, and the annotated data is used as input data to train the fault prediction model; so that the behavior pattern of future data processing equipment can be identified, providing a basis for timely handling of faults.

[0083] Based on the first normal data, multiple sample data are created, and the sample data are divided into first sample data and second sample data according to the corresponding detection unit. A first prediction model and a second prediction model are also created based on a neural network. The first sample data is used to train the first prediction model, and the second sample data is used to train the second prediction model, and the trained first prediction model and the second prediction model are obtained; the measurement values ​​between the detection units with correlation can be accurately predicted, which provides a basis for fault detection. By obtaining the measured values ​​of the first detection unit and the second detection unit, the measured values ​​of the first detection unit and the second detection unit are input into the first prediction model and the second prediction model respectively, and the predicted value of the second detection unit and the predicted value of the first detection unit are obtained respectively, the predicted value of the first detection unit is compared with the measured value of the first detection unit to obtain the first difference, and the predicted value of the second detection unit is compared with the measured value of the second detection unit to obtain the second difference, so that the state of the detection unit under the current conditions can be discovered in time, and the potential faults that may occur in the data processing device can be discovered in advance. It is also possible to obtain the prediction result of the data processing device one step earlier through mutual prediction when only the measurement value of one of the detection units is obtained. When the first difference and the second difference are respectively less than their corresponding preset thresholds, it is determined that the measured values ​​of the first detection unit and the second detection unit are accurate and annotated as normal data, and the measured values ​​of the first detection unit and the second detection unit are input into the fault prediction model to obtain the prediction result, which includes the fault type and the corresponding fault cause. Through the coordination between the above steps, the present invention can eliminate the problem of inaccurate measurement values ​​caused by damage to the detection unit in the device, thereby avoiding the adverse effects of such inaccurate measurement values ​​on fault prediction, thereby improving the accuracy of fault prediction.

[0084] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0085] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may 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. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0086] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned 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.

[0087] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0088] The above are only preferred embodiments of the present invention and are 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 in the protection scope of the present invention.

Claims

1. A method for predicting faults of a data processing device based on the Internet of Things, wherein multiple groups of sensors are installed inside or outside the data processing device, each group of sensors includes a first detection unit and a second detection unit, wherein the first detection unit and the second detection unit are sensors of different types but have correlation, based on the shape of the data processing device, the first detection unit and the second detection unit are symmetrically installed about a central axis on the data processing device, and a preset threshold is set for the difference between a measured value and a predicted value obtained by each detection unit, the detection unit refers to any one or more of the first detection unit and the second detection unit, characterized in that, The method comprises: Step S1: periodically acquiring operation data of the first detection unit and the second detection unit, and cleaning and preprocessing the operation data to acquire the preprocessed operation data, wherein the operation data is divided into first normal data and first fault data, the first normal data and the first fault data are annotated respectively, a fault prediction model is created, and the annotated first normal data and the first fault data are used as input data, and the fault type and the corresponding fault cause are used as output results of the fault prediction model to train the fault prediction model; Step S2: creating a plurality of sample data based on the first normal data, and dividing the sample data into first sample data and second sample data according to corresponding detection units, and creating a first prediction model and a second prediction model based on a neural network, training the first prediction model using the first sample data, training the second prediction model using the second sample data, and acquiring the trained first prediction model and the second prediction model; Step S3: obtaining the measured values ​​of the first detection unit and the second detection unit, inputting the measured values ​​of the first detection unit and the second detection unit into the first prediction model and the second prediction model respectively, and obtaining the predicted value of the second detection unit and the predicted value of the first detection unit respectively, comparing the predicted value of the first detection unit with the measured value of the first detection unit to obtain a first difference, and comparing the predicted value of the second detection unit with the measured value of the second detection unit to obtain a second difference; Step S4: when the first difference and the second difference are respectively less than the corresponding preset thresholds, it is determined that the measured values ​​of the first detection unit and the second detection unit are accurate, and annotated as normal data, and the measured values ​​of the first detection unit and the second detection unit are input into the fault prediction model to obtain a prediction result, which includes a fault type and a corresponding fault cause; Step S5: when the second difference is greater than or equal to the corresponding preset threshold, the measured value of the first detection unit is replaced by the predicted value of the first detection unit, and the predicted value is re-entered into the first prediction model to obtain the second predicted value of the second detection unit; when the difference between the second predicted value of the second detection unit and the measured value of the second detection unit is less than the corresponding preset threshold, it is determined that the measured value of the first detection unit is inaccurate, the measured value of the first detection unit is replaced by the predicted value of the first detection unit, and the predicted value is then combined with the measured value of the second detection unit and input into the fault prediction model to obtain a prediction result; Step S6: When the difference between the second predicted value of the second detection unit and the actual measured value of the second detection unit is greater than or equal to the corresponding preset threshold, the predicted value of the second detection unit is used to replace the actual measured value of the second detection unit, and the predicted value is re-entered into the second prediction model to obtain the second predicted value of the first detection unit. When the difference between the second predicted value of the first detection unit and the actual measured value of the first detection unit is less than the corresponding preset threshold, it is determined that the actual measured value of the second detection unit is inaccurate, and the predicted value of the second detection unit is used to replace the actual measured value of the second detection unit, and then combined with the actual measured value of the first detection unit, it is input into the fault prediction model to obtain a prediction result.

2. The method according to claim 1, characterized in that After step S6, the following steps are also included: Obtaining the measured values ​​and predicted values ​​of the detection units in other groups, and when the differences between the measured values ​​and predicted values ​​of two or more groups of detection units are greater than or equal to the corresponding preset thresholds, detecting the setting information of the data processing device, otherwise determining that the measured values ​​of the first detection unit and the second detection unit are both inaccurate; When any one or both of the measured values ​​of the first detection unit and the second detection unit are inaccurate, the corresponding measured values ​​are annotated as abnormal data and used as second abnormal data to continue training the fault prediction model.

3. The method according to claim 2, characterized in that Detecting the setting information of the data processing device includes: When the setting information of the data processing device changes, the multiple groups of measured values ​​of the detection units are annotated as normal data and used as the second normal data to retrain the first prediction model and the second prediction model. When the setting information of the data processing device does not change, the multiple groups of measured values ​​of the detection units are annotated as abnormal data and used as the second abnormal data to continue training the fault prediction model.

4. The method according to claim 1, characterized in that In step S2, dividing the sample data into first sample data and second sample data according to the corresponding detection units includes: Each of the sample data includes the measured value of the first detection unit and the measured value of the second detection unit. The measured value of the first detection unit in the sample data is used as the first sample data, and the measured value of the second detection unit in the sample data is used as the second sample data. The measured value of the first detection unit and the measured value of the second detection unit correspond to each other at a time point.

5. The method according to claim 1, characterized in that In the step S2, obtaining the trained first prediction model and the second prediction model includes: Step S21: inputting the first sample data as input data into the first prediction model, and training the first prediction model. During the training process, the first prediction model learns how to map the measured value of the first detection unit to the measured value of the second detection unit at its corresponding time point, and uses the output result of the first prediction model as the predicted value of the second detection unit; Step S22: comparing the error between the output result of the first prediction model and the actual measured value of the second detection unit, and adjusting the parameters of the first prediction model based on the error, so that the output result of the first prediction model is equal to the actual measured value of the second detection unit at the same time point corresponding to the actual measured value of the first detection unit in the first sample data; Step S23: using the actual measured value of the first detection unit and the actual measured value of the second detection unit under normal operating conditions obtained in real time as verification data, verifying the output result of the first prediction model based on the verification data, and obtaining the first prediction model when the error between the output result of the first prediction model and the actual measured value of the second detection unit is less than or equal to a set threshold; Step S24: Repeat the method from step S21 to step S23 to obtain the second prediction model.

6. The method according to claim 1, characterized in that In the step S3, obtaining the predicted value of the second detection unit and the predicted value of the first detection unit includes: The measured value of the first detection unit is input as input data into the first prediction model, and the first prediction model outputs the predicted value of the second detection unit. The measured value of the second detection unit is input as input data into the second prediction model, and the second prediction model outputs the predicted value of the first detection unit.

7. A data processing equipment fault prediction system based on the Internet of Things, used to implement the method according to any one of claims 1 to 6, characterized in that: The system includes the following modules: a data acquisition unit, configured to periodically acquire operation data of the first detection unit and the second detection unit, clean and preprocess the operation data, acquire the preprocessed operation data, divide the operation data into first normal data and first fault data, annotate the first normal data and the first fault data respectively, create a fault prediction model, and use the annotated first normal data and the first fault data as input data, and the fault type and the corresponding fault cause as output results of the fault prediction model, to train the fault prediction model; a model training unit, configured to create a plurality of sample data, and divide the sample data into first sample data and second sample data according to corresponding detection units, and further create a first prediction model and a second prediction model based on a neural network, train the first prediction model using the first sample data, train the second prediction model using the second sample data, and obtain the trained first prediction model and the second prediction model; a prediction unit, used to obtain the measured values ​​of the first detection unit and the second detection unit, input the measured values ​​of the first detection unit and the second detection unit into a first prediction model and a second prediction model respectively, and obtain the predicted value of the second detection unit and the predicted value of the first detection unit respectively, compare the predicted value of the first detection unit with the measured value of the first detection unit to obtain a first difference, and compare the predicted value of the second detection unit with the measured value of the second detection unit to obtain a second difference; a calculation unit, configured to determine that the measured values ​​of the first detection unit and the second detection unit are both accurate and annotate them as normal data when the first difference and the second difference are respectively less than the corresponding preset thresholds, input the measured values ​​of the first detection unit and the second detection unit into the fault prediction model, and obtain a prediction result, wherein the prediction result includes a fault type and a corresponding fault cause; The prediction unit is further configured to replace the actual measured value of the first detection unit with the predicted value of the first detection unit when the second difference is greater than or equal to the corresponding preset threshold, and re-input the predicted value into the first prediction model to obtain the second predicted value of the second detection unit; when the difference between the second predicted value of the second detection unit and the actual measured value of the second detection unit is less than the corresponding preset threshold, determine that the actual measured value of the first detection unit is inaccurate, replace the actual measured value of the first detection unit with the predicted value of the first detection unit, and then input the predicted value of the first detection unit together with the actual measured value of the second detection unit into the fault prediction model to obtain a prediction result; When the difference between the second predicted value of the second detection unit and the actual measured value of the second detection unit is greater than or equal to the corresponding preset threshold, the predicted value of the second detection unit is used to replace the actual measured value of the second detection unit and is re-entered into the second prediction model to obtain the second predicted value of the first detection unit. When the difference between the second predicted value of the first detection unit and the actual measured value of the first detection unit is less than the corresponding preset threshold, it is determined that the actual measured value of the second detection unit is inaccurate, and the predicted value of the second detection unit is used to replace the actual measured value of the second detection unit, and then combined with the actual measured value of the first detection unit, it is input into the fault prediction model to obtain a prediction result.

Images