Collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment

Abstract

The present invention belongs to the field of digital twin technology and discloses a collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment. The collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment performs hierarchical clustering based on the equipment type, current environmental status, environmental change law, and equipment state change law through a four-level clustering method. Through this hierarchical clustering method, it is possible to obtain reference equipment that is most similar to the operating law of the target equipment, and use the operating laws of these reference equipment to establish a deduction model to predict the next state of the target equipment. A classification state comprehensive evaluation method is also designed to evaluate the state change of the equipment, improve the reliability and stability of the equipment, and ensure the operation and safety of the Internet of Things factory.

Description

Technical Field

[0001] The present invention belongs to the field of digital twin technology, and in particular relates to a method for state deduction and evaluation of large-scale industrial equipment. Background Art

[0002] In today's digital age, the generation and accumulation of big data is growing exponentially, with the volume of data generated across all industries rapidly increasing. The widespread use of data sources such as sensors, mobile devices, and social media has led to data diversity and complexity. Traditional data processing and analysis methods face the challenge of massive data volumes, necessitating more efficient, rapid, and intelligent technologies to cope with this data deluge. Processing systems must process incoming data in real time and output analysis results promptly. As one of the core drivers of the big data era, real-time data evaluation technology not only helps companies better understand and navigate complex business environments but also improves the timeliness and accuracy of decision-making. Summary of the Invention

[0003] To address the above issues, the present invention aims to provide a collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment. This method combines digital twins with real-time data stream processing to achieve continuous monitoring and analysis of entity states. This technology obtains reference devices whose operating patterns are most similar to those of the target device and uses these operating patterns to build a deduction model to predict and evaluate the target device's next state. Furthermore, digital twins can provide contextual information and in-depth understanding for data evaluation, providing a comprehensive reference for decision makers.

[0004] In order to achieve the above object, the technical solution of the present invention is:

[0005] A collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment. The specific operating steps of the deduction and evaluation method are as follows:

[0006] Step 1: Determine the type of the target device, and then determine the type of the reference device set to obtain a clustered reference device set;

[0007] Step 2: Determine the environmental state of the target device and select a set of reference devices that operate in the same environmental state as the target device to obtain a set of bi-clustered reference devices.

[0008] Step 3: Determine the parameter change pattern of the environmental state and extract its characteristics. Further screen out the reference device set with the same environmental parameter pattern as the target device to obtain the triple cluster reference device set.

[0009] Step 4: Summarize the state change patterns of the target device itself and extract its features to perform the fourth clustering. Find reference devices with the same state change patterns as the target device to obtain a set of four-cluster reference devices.

[0010] Step 5: Based on the data types that can be collected by the equipment, they are divided into three categories, and a Bayesian network-based target equipment classification state deduction method is constructed;

[0011] Step 6: By calculating the similarity between the reference device set and the target device, the classification state deduction model of the known reference device set is integrated into the classification state deduction method of the target device;

[0012] Step 7: Comprehensively evaluate the multiple types of data derived from the target device. First, evaluate each type of data separately to ensure that they are all within the normal range. If one type of data is abnormal, the status of the target device is determined to be abnormal. Otherwise, a comprehensive evaluation is performed on multiple types of data.

[0013] Furthermore, in the above step 1, the implementation process of the re-clustering reference device set is:

[0014] Assume that there are K types of industrial equipment in the industrial Internet of Things scenario, and the type set is recorded as S:

[0015] S={S1,S2,…,S K};

[0016] The target device type is S * , confirm that the target device belongs to a specific type in the type set and record it as S v :

[0017] S * =S v , v∈[1,K];

[0018] In the industrial Internet of Things scenario, the same as the target device type, that is, all S v The devices of the same type are recorded as a clustered reference device set, the number of which is K1.

[0019] Furthermore, in the above step 2, the implementation process of the dual clustering reference device set is:

[0020] Considering the environmental factors temperature T, humidity H, and air dust content A, an environmental status evaluation formula is established to evaluate the current environmental status of the target device according to the following formula:

[0021]

[0022] Among them, E * Indicates the environmental status assessment result of the target device at the current moment. Indicates the current temperature, humidity, and dust content of the target device, T IDEA 、H IDEA 、A IDEA Indicates the temperature and humidity of the target device under ideal working conditions, T MAX 、T MIN 、H MAX 、H MIN 、A MAX 、A MIN Respectively represent the maximum and minimum values ​​of the temperature, humidity, and air dust content parameters under the working conditions of the target equipment.

[0023] And the status of each device in a clustered reference device set is evaluated according to the following formula:

[0024]

[0025] in, The environmental status evaluation result representing the environmental status evaluation result of the u1th device in a clustered reference device set at the current moment, Represents the current temperature, humidity, and air dust content of the u1th device in a clustered reference device set.

[0026] The devices in the one-fold clustering reference device set whose environmental status evaluation results are the same as those of the target device are recorded as the two-fold clustering reference device set, and the number is K2.

[0027] Furthermore, in the above step 3, the implementation process of the triple clustering reference device set is:

[0028] For the historical environmental factors of the target device, a new collection is constructed to record them:

[0029]

[0030]

[0031]

[0032] The temperature distribution in a factory environment is similar to a normal distribution, so the kurtosis and skewness of the temperature factor are calculated. Humidity and air quality are evenly distributed within a certain range. Therefore, the mean of humidity and air quality are calculated as shown in the following formula to obtain the regular characteristics of each type of environmental factor for the target equipment:

[0033]

[0034]

[0035]

[0036]

[0037] Among them, Skew(T * ) represents the direction and degree of the temperature distribution of the target device, Kurt(T * ) indicates the peak height of the temperature distribution of the target device at the average value. Indicates the average temperature of the target device, Indicates the average humidity of the target device. Represents the mean of the air quality of the target device, and constructs a new set to record the historical environmental factors of each device in the double clustering reference device set:

[0038]

[0039] The following formula is used to obtain the regular characteristics of each type of environmental factor for each device in the two-cluster reference device set:

[0040]

[0041]

[0042]

[0043]

[0044] in, Indicates the skew direction and degree of the temperature distribution of the u2th device, Indicates the peak height of the temperature distribution of the u2th device at the average value, represents the mean temperature of the u2th device, represents the mean humidity of the u2th device, The mean value of the air quality of the u2th device is compared with the regular characteristics of the corresponding environmental factors of the target device to find all devices with the same environmental change law as the target device, and obtain the triple clustering reference device set, the number of which is K3.

[0045] Furthermore, in the above step 4, the implementation process of the quadruple clustering reference device set is:

[0046] Assume that each device has m different sensors, and each sensor records n historical data. For the target device, these data are recorded in the following matrix:

[0047]

[0048] in, Indicates the nth data recorded by the mth sensor of the target device, where m and n are integers greater than or equal to 1.

[0049] Use the following formula to calculate the weight values ​​corresponding to different sensors:

[0050]

[0051] in, Indicates the ideal value of the idxth sensor when the device is working.

[0052] The following formula is used to process the data of multiple sensors at the same time to obtain the status of the target device at all historical moments:

[0053]

[0054] in, Indicates the state of the target device when recording the i-th data, and records the target device states at these historical moments in the following matrix:

[0055]

[0056] Analyze the matrix using the following formula to obtain the mean value of the state change of the target device and variance σ * :

[0057]

[0058]

[0059] For any device in the triple cluster reference device set, record its sensor data in the following matrix:

[0060]

[0061] in, represents the nth data recorded by the mth sensor of the u3th device in the triple cluster reference device set. The data of multiple sensors at the same time are processed using the following formula to obtain the status of the reference device at all historical moments:

[0062]

[0063] in, It represents the state of the u3th device in the triple cluster reference device set when recording the i-th data, and the reference device states at these historical moments are recorded in the following matrix:

[0064]

[0065] The matrix is ​​analyzed using the following formula to obtain the state change characteristics of the reference device:

[0066]

[0067] Then, the mean state change value of each device in the triple cluster reference device set is obtained and variance By comparing the state change characteristics with those of the target device, all devices with the same state change pattern as the target device are found, and a set of four-cluster reference devices is obtained, the number of which is K4.

[0068] Furthermore, in the above step 5, the target device classification state deduction method of the Bayesian network includes:

[0069] The collectible data of the equipment can be divided into three categories. The first category is data with a high correlation with time and the regularity is obviously related to time. The second category is data with a low degree of dispersion, which is fitted using Gaussian distribution. The third category is data with a high degree of dispersion that changes randomly within a specific range.

[0070] First, based on the data collected by multiple types of sensors of the devices in each quad-cluster reference device set, the data collected by the same type of sensors of all quad-cluster reference devices is used as prior knowledge and recorded in the following matrix:

[0071]

[0072] Each column of the matrix is ​​processed according to the following formula to obtain a new data sequence:

[0073]

[0074] Construct a Bayesian network and obtain the marginal probability according to the following formula:

[0075]

[0076] Based on the known marginal probabilities, calculate the conditional probabilities between adjacent data and save the conditional probabilities in the following matrix:

[0077]

[0078] Among them, d (l) To allow for a range of fluctuations, the calculated conditional probability is more representative of the collected data.

[0079] Secondly, the average conditional probability is calculated according to the following formula

[0080]

[0081] like Greater than the associated threshold probability P th , let P th =0.7, which means that there is an obvious correlation between the data at adjacent moments. The distribution characteristics of the data are further analyzed according to the following formula:

[0082]

[0083] Among them, r1 (l) Shows the degree of dispersion of the data, r1 (l) The larger it is, the more concentrated the data is. yes The interquartile range.

[0084] If r1 (l) Greater than r th , use Gaussian distribution to fit the data pattern, and further calculate the mean of the data according to the following formula and variance

[0085]

[0086]

[0087] The data follows a Gaussian distribution:

[0088]

[0089] If r1 (l) Less than r th , then the data is highly correlated with time, and the data change law model is constructed using autoregressive time series analysis according to the following formula:

[0090]

[0091] And use the least squares method to find the coefficients in the regression model

[0092] like Less than the associated threshold probability P th , it means that there is no obvious relationship between the data, that is, the conditional probability of the data between adjacent moments is low. Then the change pattern of such data belongs to the third category, which changes randomly within a certain range and has no obvious pattern:

[0093]

[0094] After obtaining the changing pattern of each type of sensor data, a specific classification state deduction model is obtained based on the data corresponding to the equipment in each reference set.

[0095] Furthermore, in step 6 above, the target device classification state deduction method includes:

[0096] The Pearson correlation coefficient is used to calculate the similarity between the target device and the reference device. The calculation formula is as follows:

[0097]

[0098] Assign a corresponding weight to each type of data in the reference device set, and then use the following formula to weightedly fuse them into a classification state deduction and evaluation model for the target device:

[0099]

[0100] Among them, f *(l) Represents the first type of state deduction method for the target device.

[0101] Furthermore, in step 7, the target device classification deduction state health comprehensive assessment method includes:

[0102] Classification deduction data for target devices First, different status assessment methods are performed according to their categories.

[0103] like For the first category of data, the classification status evaluation method is as follows:

[0104]

[0105] like For the second type of data, the classification status evaluation method is as follows:

[0106]

[0107] like For the third category of data, the classification status evaluation method is as follows:

[0108]

[0109]

[0110] If the evaluation results of each classification status meet the following formula, the classification status evaluation results are all within the normal range:

[0111]

[0112] in, They respectively represent the normal evaluation lower limit value and normal evaluation upper limit value of the lth classification state of the target device.

[0113] And further conduct a comprehensive status assessment of the target:

[0114]

[0115] If the comprehensive status satisfies the following formula, the target devices are in normal status:

[0116]

[0117] in, They respectively represent the lowest normal evaluation value and the highest normal evaluation value of the comprehensive status of the target device.

[0118] In summary, the present invention has the following beneficial effects:

[0119] 1. The deduction and evaluation method described in this invention uses a four-level clustering approach to perform hierarchical clustering based on device type, current environmental state, environmental change patterns, and device state change patterns. This hierarchical clustering method can identify reference devices whose operating patterns are most similar to those of the target device. Using these reference devices' operating patterns, a deduction model is then built to predict the target device's next state.

[0120] 2. A comprehensive classification status evaluation method was further designed to evaluate the status changes of equipment, improve the reliability and stability of equipment, and ensure the operation and safety of IoT factories.

[0121] 3. Compared with traditional data processing and analysis methods, the deduction and evaluation method of the present invention is more efficient, fast and intelligent. It can process the received data in real time and output the analysis results in a timely manner, helping enterprises better understand and cope with complex business environments, and also improving the timeliness and accuracy of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0122] Figure 1 is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0123] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0124] Example 1

[0125] This embodiment relates to a collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment, such as Figure 1 The specific steps are as follows:

[0126] Step S1: Determine the type of the target device, and then determine the type of the reference device set to obtain a re-clustered reference device set.

[0127] Assume that there are K types of industrial equipment in the industrial Internet of Things scenario, and the type set is recorded as S:

[0128] S={S1,S2,…,S K};

[0129] The target device type is S * , confirm that the target device belongs to a specific type in the type set and record it as S v :

[0130] S * =S v , v∈[1,K];

[0131] In the industrial Internet of Things scenario, the same as the target device type, that is, all S v The devices of the same type are recorded as a clustered reference device set, the number of which is K1.

[0132] Step S2: Determine the environmental state of the target device, and select a set of reference devices that operate in the same environmental state as the target device to obtain a double-clustered reference device set.

[0133] Considering the environmental factors temperature T, humidity H, and air dust content A, an environmental status evaluation formula is established to evaluate the current environmental status of the target device according to the following formula:

[0134]

[0135] Among them, E * Indicates the environmental status assessment result of the target device at the current moment. Indicates the current temperature, humidity, and dust content of the target device, T IDEA 、H IDEA 、A IDEA Indicates the temperature and humidity of the target device under ideal working conditions, T MAX 、T MIN 、H MAX 、H MIN 、A MAX 、A MIN Respectively represent the maximum and minimum values ​​of the temperature, humidity, and air dust content parameters under the working conditions of the target equipment.

[0136] And the status of each device in a clustered reference device set is evaluated according to the following formula:

[0137]

[0138] in, The environmental status evaluation result representing the environmental status evaluation result of the u1th device in a clustered reference device set at the current moment, Represents the current temperature, humidity, and air dust content of the u1th device in a clustered reference device set.

[0139] The devices in the one-cluster reference device set whose environmental status evaluation results are the same as those of the target device are recorded as a two-cluster reference device set, and the number is K2.

[0140] Step S3: Determine the parameter change pattern of the environmental state and extract its characteristics, further filter out the reference device set with the same environmental parameter pattern as the target device, and obtain the triple cluster reference device set:

[0141] For the historical environmental factors of the target device, a new collection is constructed to record them:

[0142]

[0143]

[0144]

[0145] The temperature distribution in a factory environment is similar to a normal distribution, so the kurtosis and skewness of the temperature factor are calculated. Humidity and air quality are evenly distributed within a certain range. Therefore, the mean of humidity and air quality are calculated as shown in the following formula to obtain the regular characteristics of each type of environmental factor for the target equipment:

[0146]

[0147]

[0148]

[0149]

[0150] Among them, Skew(T * ) represents the direction and degree of the temperature distribution of the target device, Kurt(T * ) indicates the peak height of the temperature distribution of the target device at the average value. Indicates the average temperature of the target device, Indicates the average humidity of the target device. Represents the mean of the air quality of the target device, and constructs a new set to record the historical environmental factors of each device in the double clustering reference device set:

[0151]

[0152] The following formula is used to obtain the regular characteristics of each type of environmental factor for each device in the two-cluster reference device set:

[0153]

[0154]

[0155]

[0156]

[0157] in, Indicates the skew direction and degree of the temperature distribution of the u2th device, Indicates the peak height of the temperature distribution of the u2th device at the average value, represents the mean temperature of the u2th device, represents the mean humidity of the u2th device, The mean value of the air quality of the u2th device is compared with the regular characteristics of the corresponding environmental factors of the target device to find all devices with the same environmental change law as the target device, and obtain the triple clustering reference device set, the number of which is K3.

[0158] Step S4: Summarize the state change rules of the target device itself, extract its features, perform the fourth clustering, find reference devices with the same state change rules as the target device, and obtain a set of quadruple clustered reference devices.

[0159] Assume that each device has m different sensors, and each sensor records n historical data. For the target device, these data are recorded in the following matrix:

[0160]

[0161] in, Indicates the nth data recorded by the mth sensor of the target device.

[0162] Use the following formula to calculate the weight values ​​corresponding to different sensors:

[0163]

[0164] in, Indicates the ideal value of the idxth sensor when the device is working.

[0165] The following formula is used to process the data of multiple sensors at the same time to obtain the status of the target device at all historical moments:

[0166]

[0167] in, Indicates the state of the target device when recording the i-th data, and records the target device states at these historical moments in the following matrix:

[0168]

[0169] Analyze the matrix using the following formula to obtain the mean value of the state change of the target device and variance σ * :

[0170]

[0171]

[0172] For any device in the triple cluster reference device set, record its sensor data in the following matrix:

[0173]

[0174] in, represents the nth data recorded by the mth sensor of the u3th device in the triple cluster reference device set. The data of multiple sensors at the same time are processed using the following formula to obtain the status of the reference device at all historical moments:

[0175]

[0176] in, It represents the state of the u3th device in the triple cluster reference device set when recording the i-th data, and the reference device states at these historical moments are recorded in the following matrix:

[0177]

[0178] The matrix is ​​analyzed using the following formula to obtain the state change characteristics of the reference device:

[0179]

[0180] Then, the mean state change value of each device in the triple cluster reference device set is obtained and variance By comparing the state change characteristics with those of the target device, all devices with the same state change pattern as the target device are found, and a set of four-cluster reference devices is obtained, the number of which is K4.

[0181] Step S5 , by calculating the similarity between the reference device set and the target device, the classification state deduction model of the known reference device set is integrated into the classification state deduction method of the target device.

[0182] The collectible data of the equipment can be divided into three categories. The first category is data with a high correlation with time and the regularity is obviously related to time. The second category is data with a low degree of dispersion, which is fitted using Gaussian distribution. The third category is data with a high degree of dispersion that changes randomly within a specific range.

[0183] Step S5.1: Based on the data collected by multiple types of sensors of the devices in each quad-cluster reference device set, the data collected by the same type of sensors of all quad-cluster reference devices are used as prior knowledge and recorded in the following matrix:

[0184]

[0185] Each column of the matrix is ​​processed according to the following formula to obtain a new data sequence:

[0186]

[0187] Construct a Bayesian network and obtain the marginal probability according to the following formula:

[0188]

[0189] Based on the known marginal probabilities, calculate the conditional probabilities between adjacent data and save the conditional probabilities in the following matrix:

[0190]

[0191] Among them, d (l) To allow for a range of fluctuations, the calculated conditional probability is more representative of the collected data.

[0192] Step S5.2, calculate the average conditional probability according to the following formula

[0193]

[0194] (1) If Greater than the associated threshold probability P th , let P th =0.7, which means that there is an obvious correlation between the data at adjacent moments. The distribution characteristics of the data are further analyzed according to the following formula:

[0195]

[0196] Among them, r1 (l) Shows the degree of dispersion of the data, r1 (l) The larger it is, the more concentrated the data is. yes The interquartile range.

[0197] (a) If r1(l) Greater than r th , use Gaussian distribution to fit the data pattern, and further calculate the mean of the data according to the following formula and variance

[0198]

[0199]

[0200] The data follows a Gaussian distribution:

[0201]

[0202] (b) If r1 (l) Less than r th , then the data is highly correlated with time, and the data change law model is constructed using autoregressive time series analysis according to the following formula:

[0203]

[0204] And use the least squares method to find the coefficients in the regression model

[0205] (2) If Less than the associated threshold probability P th , it means that there is no obvious relationship between the data, that is, the conditional probability of the data between adjacent moments is low. Then the change pattern of such data belongs to the third category, which changes randomly within a certain range and has no obvious pattern:

[0206]

[0207] After obtaining the changing pattern of each type of sensor data, a specific classification state deduction model is obtained based on the data corresponding to the equipment in each reference set.

[0208] Step S6: By calculating the similarity between the reference device set and the target device, the classification state deduction model of the known reference device set is integrated into the classification state deduction method of the target device.

[0209] The Pearson correlation coefficient is used to calculate the similarity between the target device and the reference device. The calculation formula is as follows:

[0210]

[0211] Assign a corresponding weight to each type of data in the reference device set, and then use the following formula to weightedly fuse them into a classification state deduction and evaluation model for the target device:

[0212]

[0213] Among them, f *(l) Represents the first type of state deduction method for the target device.

[0214] Step S7, comprehensive evaluation of multiple types of data derived from the target device. First, each type of data is evaluated separately to determine whether they are all within the normal range. If one type of data is abnormal, the status of the target device is determined to be abnormal. Otherwise, a comprehensive evaluation is performed on multiple types of data.

[0215] Classification deduction data for target devices First, different status assessment methods are performed according to their categories.

[0216] like For the first category of data, the classification status evaluation method is as follows:

[0217]

[0218] like For the second type of data, the classification status evaluation method is as follows:

[0219]

[0220] like For the third category of data, the classification status evaluation method is as follows:

[0221]

[0222]

[0223] If the evaluation results of each classification status meet the following formula, the classification status evaluation results are all within the normal range:

[0224]

[0225] in, They respectively represent the normal evaluation lower limit value and normal evaluation upper limit value of the lth classification state of the target device.

[0226] And further conduct a comprehensive status assessment of the target:

[0227]

[0228]

[0229] If the comprehensive status satisfies the following formula, the target devices are in normal status:

[0230]

[0231] in, It should be noted that the above embodiments are only intended to help understand the method and core concept of this application. For those skilled in the art, without departing from the principles of this application, several improvements and modifications can be made to this application, and these improvements and modifications are also within the scope of protection of the claims of this application.

Claims

1. A collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment, characterized by: The specific operation steps of the deduction and evaluation method are as follows: Step 1: Determine the type of the target device, and then determine the type of the reference device set to obtain a clustered reference device set; Step 2: Determine the environmental state of the target device and select a set of reference devices that operate in the same environmental state as the target device to obtain a set of bi-clustered reference devices. Step 3: Determine the parameter change pattern of the environmental state and extract its characteristics. Further screen out the reference device set with the same environmental parameter pattern as the target device to obtain the triple cluster reference device set. Step 4: Summarize the state change patterns of the target device itself and extract its features to perform the fourth clustering. Find reference devices with the same state change patterns as the target device to obtain a set of four-cluster reference devices. Step 5: Based on the data types that can be collected by the equipment, they are divided into three categories, and a Bayesian network-based target equipment classification state deduction method is constructed; Step 6: By calculating the similarity between the reference device set and the target device, the classification state deduction model of the known reference device set is integrated into the classification state deduction method of the target device; Step 7: Comprehensively evaluate the multiple types of data derived from the target device. First, evaluate each type of data separately to ensure that they are all within the normal range. If one type of data is abnormal, the status of the target device is determined to be abnormal. Otherwise, a comprehensive evaluation is performed on multiple types of data.

2. The collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment according to claim 1 is characterized in that: The implementation process of re-clustering the reference device set in step 1 is as follows: Assume that there are K types of industrial equipment in the industrial Internet of Things scenario, and the type set is recorded as S: S={S1,S2,…,S K }; The target device type is S * , confirm that the target device belongs to a specific type in the type set and record it as S v : S * =S v ,v∈[1,K]; The device type is used as the clustering sample parameter. In the industrial Internet of Things scenario, the same type as the target device, that is, all devices belonging to S v The devices of the same type are recorded as a clustered reference device set, the number of which is K1.

3. The collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment according to claim 1 is characterized in that: The implementation process of the double clustering reference device set in step 2 is: Considering the environmental factors temperature T, humidity H, and air dust content A, an environmental status evaluation formula is established to evaluate the current environmental status of the target device according to the following formula: Among them, E * Indicates the environmental status assessment result of the target device at the current moment. Indicates the current temperature, humidity, and dust content of the target device, T IDEA 、H IDEA 、A IDEA Indicates the temperature and humidity of the target device under ideal working conditions, T MAX 、T MIN 、H MAX 、H MIN 、A MAX 、A MIN Respectively represent the maximum and minimum values ​​of the temperature, humidity and air dust content parameters under the working conditions of the target equipment; And the status of each device in a clustered reference device set is evaluated according to the following formula: in, The environmental status evaluation result representing the environmental status evaluation result of the u1th device in a clustered reference device set at the current moment, represents the current temperature, humidity, and air dust content of the u1th device in a clustered reference device set; The state evaluation value is used as the clustering sample parameter. The devices in the single-cluster reference device set whose environmental state evaluation results are the same as those of the target device are recorded as the double-cluster reference device set, and the number is K2.

4. The collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment according to claim 1 is characterized in that: The implementation process of the triple clustering reference device set in step 3 is as follows: For the historical environmental factors of the target device, a new collection is constructed to record them: The temperature distribution in a factory environment is similar to a normal distribution, so the kurtosis and skewness of the temperature factor are calculated. Humidity and air quality are evenly distributed within a certain range. Therefore, the mean of humidity and air quality are calculated as shown in the following formula to obtain the regular characteristics of each type of environmental factor for the target equipment: Among them, Skew(T * ) represents the direction and degree of the temperature distribution of the target device, Kurt(T * ) indicates the peak height of the temperature distribution of the target device at the average value. Indicates the average temperature of the target device, Indicates the average humidity of the target device. Represents the mean of the air quality of the target device, and constructs a new set to record the historical environmental factors of each device in the double clustering reference device set: The following formula is used to obtain the regular characteristics of each type of environmental factor for each device in the two-cluster reference device set: in, Indicates the skew direction and degree of the temperature distribution of the u2th device, Indicates the peak height of the temperature distribution of the u2th device at the average value, represents the mean temperature of the u2th device, Indicates the average humidity of the environment where the u2th device is located. The mean value of the ambient air quality of the u2th device is compared with the regular characteristics of the corresponding environmental factors of the target device. The skew direction and degree of temperature distribution, the peak height of the temperature distribution at the average value, the mean value of the ambient humidity of the device, and the mean value of the ambient air quality of the device are used as clustering sample parameters to find all devices with the same environmental change law as the target device, and obtain a triple cluster reference device set, the number of which is K3.

5. The collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment according to claim 1 is characterized in that: The implementation process of the quadruple clustering reference device set in step 4 is as follows: Assume that each device has m different sensors, and each sensor records n historical data. For the target device, these data are recorded in the following matrix: in, Indicates the nth data recorded by the mth sensor of the target device, where m and n are integers greater than or equal to 1; Use the following formula to calculate the weight values ​​corresponding to different sensors: in, Indicates the ideal value of the idxth sensor when the device is working; The following formula is used to process the data of multiple sensors at the same time to obtain the status of the target device at all historical moments: in, Indicates the state of the target device when recording the i-th data, and records the target device states at these historical moments in the following matrix: Analyze the matrix using the following formula to obtain the mean value of the state change of the target device: and variance σ * : For any device in the triple cluster reference device set, record its sensor data in the following matrix: in, The nth data recorded by the mth sensor of the u3th device in the triple cluster reference device set is represented. The data of multiple sensors at the same time are processed using the following formula to obtain the status of the reference device at all historical moments: in, It represents the state of the u3th device in the triple cluster reference device set when recording the i-th data, and the reference device states at these historical moments are recorded in the following matrix: The matrix is ​​analyzed using the following formula to obtain the state change characteristics of the reference device: Then, the mean state change value of each device in the triple cluster reference device set is obtained and variance Compare the state change characteristics with those of the target device, use the state change mean and variance of the device as clustering sample parameters, find all devices with the same state change pattern as the target device, and obtain a set of four-fold cluster reference devices, the number of which is K4.

6. The collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment according to claim 1 is characterized in that: The target device classification state deduction method of the Bayesian network in step 5 above includes: The collectible data of the equipment is divided into three categories. The first category is data with a high correlation with time and a clear regularity with time. The second category is data with a low degree of dispersion, which is fitted using Gaussian distribution. The third category is data with a high degree of dispersion that changes randomly within a specific range. First, based on the data collected by multiple types of sensors of the devices in each quad-cluster reference device set, the data collected by the same type of sensors of all quad-cluster reference devices is used as prior knowledge and recorded in the following matrix: Each column of the matrix is ​​processed according to the following formula to obtain a new data sequence: Construct a Bayesian network and obtain the marginal probability according to the following formula: Based on the known marginal probabilities, calculate the conditional probabilities between adjacent data and save the conditional probabilities in the following matrix: Among them, d (l) To allow for a range of fluctuations so that the calculated conditional probability is more representative of the collected data; Secondly, the average conditional probability is calculated according to the following formula like Greater than the associated threshold probability P th , let P th =0.7, which means that there is an obvious correlation between the data at adjacent moments. The distribution characteristics of the data are further analyzed according to the following formula: in, It shows the degree of dispersion of the data. The larger it is, the more concentrated the data is. yes The interquartile range of Greater than r th , use Gaussian distribution to fit the data pattern, and further calculate the mean of the data according to the following formula and variance The data follows a Gaussian distribution: like Less than r th , then the data is highly correlated with time, and the data change law model is constructed using autoregressive time series analysis according to the following formula: And use the least squares method to find the coefficients in the regression model like Less than the associated threshold probability P th , it means that there is no obvious relationship between the data, that is, the conditional probability of the data between adjacent moments is low. Then the change pattern of such data belongs to the third category, which changes randomly within a certain range and has no obvious pattern: After obtaining the changing pattern of each type of sensor data, a specific classification state deduction model is obtained based on the data corresponding to each device in the reference set.

7. The collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment according to claim 1 is characterized in that: The target device classification state deduction method in step 6 includes: The Pearson correlation coefficient is used to calculate the similarity between the target device and the reference device. The calculation formula is as follows: Assign a corresponding weight to each type of data in the reference device set, and then use the following formula to weightedly fuse them into a classification state deduction model for the target device: Among them, f *(l) Represents the first type of state deduction method for the target device.

8. The collaborative comparative digital twin deduction and evaluation method for large-scale industrial equipment according to claim 1 is characterized in that: The target device classification deduction state health comprehensive assessment method in step 7 includes: Classification deduction data for target devices First, different status assessment methods are used according to their categories; like For the first category of data, the classification status evaluation method is as follows: like For the second type of data, the classification status evaluation method is as follows: like For the third category of data, the classification status evaluation method is as follows: If the evaluation results of each classification status meet the following formula, the classification status evaluation results are all within the normal range: in, They represent the lower limit and upper limit of normal evaluation of the target device's lth classification state respectively; and further perform a comprehensive state evaluation on the target: If the comprehensive status satisfies the following formula, the target devices are in normal status: in, They respectively represent the lowest normal evaluation value and the highest normal evaluation value of the comprehensive status of the target device.

Images