CN110008565B - Prediction method of abnormal working condition of industrial process based on correlation analysis of operating parameters - Google Patents

Abstract

The invention discloses an industrial process abnormal working condition prediction method based on operation parameter correlation analysis, which can be applied to fault prediction and health management of an industrial process. The method starts from the relevance among the operation parameters of the industrial process, and carries out the prediction of the abnormal working condition based on the relevance analysis of the operation parameters. In the single-parameter prediction stage, the method predicts each operation parameter through an exponential smoothing method according to the existing sensor data. In the correlation analysis stage, the invention calculates the correlation of the operation parameters by the known values of the operation parameters and the predicted values of the parameters, wherein the correlation of the parameters is represented by the similarity of a series of indexes representing the parameter curves. In the relevance trend prediction stage, the method constructs a multiple autoregressive model to predict the parameter relevance. The method provided by the invention takes the relevance of the operation parameters into consideration, can obtain more complete equipment abnormal information and more advanced prediction results, and has practical significance for the fault prediction of industrial equipment.

Description

Prediction method of abnormal working condition of industrial process based on correlation analysis of operating parameters

technical field

The invention belongs to the technical field of reliability engineering, and relates to a method for predicting abnormal working conditions of an industrial process based on correlation analysis of operating parameters.

Background technique

With the continuous emergence of complex systems and the increasing demand for real-time monitoring of industrial processes, modern industrial equipment is often equipped with multiple sensors to monitor its operating status during operation. At the same time, during the operation of the equipment, there may be various failure modes, and a certain failure may correspond to several symptoms. In this case, the information of a single sensor can no longer fully reflect the operation status of the equipment, and the failure prediction based on multi-sensor information emerges as the times require. Fault prediction based on multi-sensor information aims to analyze the operating status of equipment using comprehensive sensor information, thereby enabling more reliable equipment diagnosis and prediction. With the continuous development of sensing technology, the use of multiple sensors for equipment condition monitoring, fault diagnosis and prediction has become a development trend.

For multiple sensors during the operation of the equipment, the operating parameters represented by them do not exist independently. The change of each operating parameter during the operation of the equipment is actually a response to the current operating state of the equipment. In the normal operation state of industrial equipment, the operating parameters are usually relatively stable and maintained at a relatively stable level, so the correlation of the operating parameters is relatively stable. However, when abnormal working conditions occur, the response of each operating parameter to the abnormality is different, which leads to the change of the correlation of the operating parameters.

SUMMARY OF THE INVENTION

In view of the existing technical situation, the purpose of the present invention is to solve the problem of correlation between various operating parameters during the operation of the equipment, to judge the state of the equipment operating process through the correlation of the operating parameters, and to determine the trend of the correlation of the operating parameters. Prediction of abnormal working conditions of equipment.

The concept of the present invention is now described as follows:

The invention proposes a method for predicting abnormal working conditions of an industrial process based on the change trend of the correlation of operating parameters, starting from the correlation between the operating parameters of the industrial process, and predicting the abnormal working conditions through trend analysis of the correlation of the operating parameters. In the single-parameter prediction stage, the present invention predicts each operating parameter through the exponential smoothing method according to the existing sensor data. In the correlation analysis stage, the present invention calculates the correlation of the operating parameters through the known values of each operating parameter and the predicted value of the parameters, wherein the correlation of the parameters is represented by the similarity of a series of indicators representing the parameter curve. In the correlation trend prediction stage, the present invention constructs a multivariate autoregressive model to predict the parameter correlation. The method proposed by the present invention takes the correlation of the operating parameters into consideration, and can obtain more complete equipment abnormality information and a more advanced prediction result.

According to the above inventive concept, the present invention proposes a method for predicting abnormal working conditions of an industrial process based on correlation analysis of operating parameters, which includes the following steps:

Step 1: Use the Holt exponential smoothing model to predict the sequence of measurement values collected by each sensor in the industrial process with a fixed step size;

Step 2: Construct triples characterized by the changing trend of industrial process operating parameters in the form of sliding windows, represent the measured value sequence in the data window, and calculate any arbitrary value in the window through the correlation index based on Euclidean distance. The correlation of two operating parameters;

Step 3: construct a relevance prediction model according to the calculated relevance data, the relevance prediction model is a multiple autoregressive model, and the model parameters are estimated by a partial least squares algorithm;

Step 4: For the newly obtained data, predict the abnormal working conditions of the equipment according to the correlation prediction model, and continuously update the correlation prediction model and model parameters before the occurrence of abnormal working conditions is predicted.

Based on the above solution, each step can be implemented in the following manner:

Preferably, step 1 is as follows:

其中K表示序列长度，

表示传感器i在第k个采样时刻点的测量值；Step 1.1: For industrial equipment with multiple sensors, record the number of sensors as N. When the equipment is in operation, the operating parameter values that characterize the operating state of the equipment, that is, sensor data, are continuously collected, and each sensor measurement sequence is recorded as

where K is the sequence length,

represents the measurement value of sensor i at the kth sampling time point;

其平滑值可根据下式计算：Step 1.2: For sensor i, use the Holt exponential smoothing model to predict its measurement sequence, given the measurement

Its smoothed value can be calculated according to the following formula:

表示

的平滑值；

是线性增长因子，代表平滑后的趋势；α和β是平滑系数，取值范围均为(0，1)；Holt模型的初始条件如下：in,

express

the smoothed value of ;

is a linear growth factor, representing the trend after smoothing; α and β are smoothing coefficients, both in the range of (0, 1); the initial conditions of the Holt model are as follows:

为：Step 1.3: After obtaining the smoothed value of the measured data and the linear growth factor, use the result to predict, predict the value

for:

where l represents the prediction step size and τ is the sampling interval of the device sensor signal.

Preferably, step 2 includes the following sub-steps:

其中j为该窗口起点，对应的采样时刻为t_j；若假设该窗口内其中一个时间段内的m个数据

能够由一条线段进行拟合，对于其中的测量值

其在拟合线段中所对应的值为

则该线段的拟合误差ERR计算公式为：Step 2.1: Running parameter correlation analysis is divided into three stages: time series segmental fitting, triple representation and correlation calculation; in the time series segmental fitting stage, for a data window with a fixed length of L, it is recorded as a window W _j , running parameters X ⁱ , i=1, 2,...,N, the L data in this window are

where j is the starting point of the window, and the corresponding sampling time is t _j ; if it is assumed that m data in one of the time periods in the window

can be fitted by a line segment, for the measured values in

Its corresponding value in the fitted line segment is

Then the calculation formula of the fitting error ERR of the line segment is:

开始对其进行线段拟合；在数据分段拟合的过程按以下步骤进行，其中记设定的拟合误差阈值为ω_E：When performing piecewise linearization on the window data starting from j, from

Begin to perform line segment fitting on it; in the process of data segment fitting, follow the steps below, where the set fitting error threshold is ω _E :

拟合终点为

其中，对于以j为起点的窗口W_j，初始的拟合起点为

Step (1): Set the fitting starting point as

The fitting end point is

Among them, for the window W _j with j as the starting point, the initial fitting starting point is

采用直线回归的方式进行线段拟合，由此获得相应的线段数据

根据所述的拟合误差ERR计算公式计算其拟合误差ERR；Step (2): For the data

Line segment fitting is performed by means of linear regression, thereby obtaining the corresponding line segment data

Calculate its fitting error ERR according to the fitting error ERR calculation formula;

Step (3): If ERR≤ω _E , set h=h+1, and repeat step (2); if ERR>ω _E , save the current fitting end point (ie, the data split point), and reset h=2 , and go back to step 1, and use the current fitting end point as the new fitting starting point to fit the next part of the data;

Repeat the above steps (1) to (3) until all the data in the window W _j have been segmented, that is, the piecewise linearized data is obtained

Step 2.2: In the line segment triple representation stage, a line segment s _j is described in the form of the following triples,

表示该线段在时间轴上的长度，r_j表示该线段数值的增长率，即对于线段数据

对于窗口W_j数据线段化后的新序列

得到其三元组序列表示形式为{s₁，s₂，...，s_n}，其中n表示该窗口内数据分段后的线段数量；where k _j represents the slope of the line segment,

Represents the length of the line segment on the time axis, and r _j represents the growth rate of the line segment value, that is, for the line segment data

For the new sequence of window W _j data segmented

Obtain its triplet sequence representation as {s ₁ , s ₂ , ..., s _n }, where n represents the number of line segments after data segmentation in the window;

参数V^B的分段点为

其中n_A和n_B分别表示运行参数A和B的测量数据在该窗口内分段后的线段数；对参数V^A和V^B的分段点进行合并，去除重复项并进行从小到大排列，得到两个参数的分割点序列为

随后，根据该分割点序列以及三元组表示形式获得参数V^A和V^B新的三元组序列为

和

Step 2.3: In the correlation calculation stage, for the two operating parameters VA and ^VB during the operation of the equipment, the segmented data needs to be segmented first: in the window ^W _j , record the score of the parameter VA ^. segment point is

The segmentation point of parameter ^VB is

Among them, n _A and n _B respectively represent the number of line segments of the measurement data of operating parameters ^A and B in this window; the segment points of parameters VA and ^VB are merged, duplicates are removed, and they are arranged from small to large , the split point sequence of the two parameters is obtained as

Then, the parameters VA and ^VB are obtained according to the split point sequence and the triplet representation ^. The new triplet sequence is

and

After obtaining the triplet sequence of the two parameters VA and ^VB in the window ^W _j , the correlation between the two parameters VA and ^VB in the window is calculated by the correlation index d ^AB based on the Euclidean distance ^:

为参数V^A第i条线段的斜率，

为参数V^B第i条线段的斜率，

为参数V^A第i条线段的数值增长率，

为参数V^B第i条线段的数值增长率。where:

is the slope of the ith line segment of parameter V ^A ,

is the slope of the ith line segment of parameter V ^B ,

is the numerical growth rate of the ith line segment of parameter V ^A ,

is the numerical growth rate of the ith line segment of parameter V ^B.

Preferably, step 3 includes the following sub-steps:

Step 3.1: Construct the correlation prediction model: set the prediction step size to f, determine the model design parameters U and M according to the known data length so that U+f+M-1=the total number of windows W _j , and construct the following matrix:

Among them, {d ₁ , d ₂ , ..., d _U+f+M-1 ^} represents the correlation sequence of the parameters VA and ^VB , and d ^j represents the two parameters VA and _VB in the window ^W _j Relevance index d ^AB ;

Step 3.2: For each correlation sequence, use the U correlation values to predict the correlation value after f step, and construct a correlation prediction model:

F _p =D _p θ

The parameters θ=[θ ₁ , θ ₂ , . . . , θ _U ] ^T can be obtained by partial least squares algorithm.

Preferably, step 4 includes the following sub-steps:

When predicting, construct a new matrix with the newly acquired data from the sensor:

Subsequently, the correlation value is predicted in f-steps using the constructed correlation prediction model, i.e.

判断设备出现异常，其中d_normal为设备运行初期处于正常运行状态时参数V^A和V^B的关联性值，ω_p为关联性值相对于初始正常的漂移量阈值；若

则利用传感器获得的新数据重构模型以更新模型参数θ；when

It is judged that the equipment is abnormal, where d _normal is the correlation value of the parameters VA and ^VB when the equipment is in ^a normal operating state in the early stage of operation, and ω _p is the drift threshold of the correlation value relative to the initial normal; if

Then use the new data obtained by the sensor to reconstruct the model to update the model parameter θ;

With the update of the data, the correlation value of a certain prediction step is continuously predicted, so as to predict the occurrence time of abnormal working conditions of the equipment.

The method for predicting abnormal working conditions of an industrial process based on the correlation analysis of the operating parameters proposed by the invention can be used for a complex industrial system with multiple sensors. The invention starts from the correlation between the operating parameters of the industrial process, and predicts abnormal working conditions based on the correlation analysis of the operating parameters. In the single-parameter prediction stage, the present invention predicts each operating parameter through the exponential smoothing method according to the existing sensor data. In the correlation analysis stage, the present invention calculates the correlation of the operating parameters through the known values of each operating parameter and the predicted value of the parameters, wherein the correlation of the parameters is represented by the similarity of a series of indicators representing the parameter curve. In the correlation trend prediction stage, the present invention constructs a multivariate autoregressive model to predict the parameter correlation. The method proposed by the present invention takes the correlation of the operating parameters into consideration, and can obtain more complete equipment abnormality information and a more advanced prediction result. This will provide strong data support for subsequent equipment health management, which is especially valuable for high-reliability equipment maintenance management, and has broad prospects in practical engineering applications.

Description of drawings

Figure 1. The measured values and prediction results of steam turbine operating parameters;

Figure 2. Comparison of the trend prediction results of the correlation between the vacuum A of the steam turbine and other parameters and the true value;

Fig. 3 Prediction results of turbine abnormality occurrence time.

Detailed ways

The specific embodiments of the present invention will now be further described with reference to the accompanying drawings. Part of the principles have been described in detail above and will not be repeated here. The following example uses a real case based on the low vacuum protection trip data of the steam turbine to illustrate the specific operation steps and verify the effectiveness of the proposed method.

This milling machine data records the running degradation process of cutting metal materials with milling cutters. The initial operating condition of the steam turbine is the load of 250MW and the vacuum of the condenser of 93kPa. Taking the vacuum value of the condenser as the indicator parameter, the vacuum A starts to indicate abnormality from the 762nd sampling point. When the vacuum value drops to 81kPa, the steam turbine jumps machine. The method for predicting abnormal working conditions in an industrial process includes the following steps:

Step 1: Use the Holt exponential smoothing model to predict the sequence of measurement values collected by each sensor in the industrial process with a fixed step size. This step specifically includes the following sub-steps:

其中K表示序列长度，

表示传感器i在第k个采样时刻点的测量值；Step 1.1: For industrial equipment with multiple sensors, record the number of sensors as N. When the equipment is in operation, the operating parameter values that characterize the operating state of the equipment, that is, sensor data, are continuously collected, and each sensor measurement sequence is recorded as

where K is the sequence length,

represents the measurement value of sensor i at the kth sampling time point;

其平滑值可根据下式计算：Step 1.2: For sensor i, use the Holt exponential smoothing model to predict its measurement sequence, given the measurement

Its smoothed value can be calculated according to the following formula:

表示

的平滑值；

是线性增长因子，代表平滑后的趋势；α和β是平滑系数，取值范围均为(0，1)；Holt模型的初始条件如下：in,

express

the smoothed value of ;

is a linear growth factor, representing the trend after smoothing; α and β are smoothing coefficients, both in the range of (0, 1); the initial conditions of the Holt model are as follows:

为：Step 1.3: After obtaining the smoothed value of the measured data and the linear growth factor, use the result to predict, predict the value

for:

where l represents the prediction step size and τ is the sampling interval of the device sensor signal. In this example, the preset prediction step size is l=15.

According to step 1, for each operating parameter measurement sequence, a fixed-step prediction is performed, and the result is given in Figure 1, and the actual condition monitoring measurement sequence is also given at the same time.

Step 2: Construct triples characterized by the changing trend of industrial process operating parameters in the form of sliding windows, represent the measured value sequence in the data window, and calculate any arbitrary value in the window through the correlation index based on Euclidean distance. Dependency of two run parameters. This step specifically includes the following sub-steps:

其中j为该窗口起点，对应的采样时刻为t_j；若假设该窗口内其中一个时间段内的m个数据

恰好能够由一条线段进行拟合，对于其中的测量值

其在拟合线段中所对应的值为

则该线段的拟合误差ERR计算公式为：Step 2.1: Running parameter correlation analysis is divided into three stages: time series segmental fitting, triple representation and correlation calculation; in the time series segmental fitting stage, for a data window with a fixed length of L, it is recorded as a window W _j , in this case, the sliding window length is 100. Running parameters X ⁱ , i=1, 2,...,N, the L data in this window are

where j is the starting point of the window, and the corresponding sampling time is t _j ; if it is assumed that m data in one of the time periods in the window

can be fitted by exactly one line segment, for the measured values in

Its corresponding value in the fitted line segment is

Then the calculation formula of the fitting error ERR of the line segment is:

开始对其进行线段拟合；在数据分段拟合的过程按以下步骤进行(其中记设定的拟合误差阈值为ω_E，本例中ω_E＝0.025)：When performing piecewise linearization on the window data starting from j, from

Begin to perform line segment fitting on it; in the process of data segment fitting, follow the following steps (where the set fitting error threshold is ω _E , in this example ω _E =0.025):

拟合终点为

其中，对于以j为起点的窗口W_j，初始的拟合起点为

Step (1): Set the fitting starting point as

The fitting end point is

Among them, for the window W _j with j as the starting point, the initial fitting starting point is

采用直线回归的方式进行线段拟合，由此获得相应的线段数据

根据所述的拟合误差ERR计算公式计算其拟合误差ERR；Step (2): For the data

Line segment fitting is performed by means of linear regression, thereby obtaining the corresponding line segment data

Calculate its fitting error ERR according to the fitting error ERR calculation formula;

Step (3): If ERR≤ω _E , set h=h+1, and repeat step (2); if ERR>ω _E , save the current fitting end point (ie, the data split point), and reset h=2 , and go back to step 1, and use the current fitting end point as the new fitting starting point to fit the next part of the data;

Repeat the above steps (1) to (3) until all the data in the window W _j have been segmented, that is, the piecewise linearized data is obtained

Step 2.2: In the line segment triple representation stage, a line segment s _j is described in the form of the following triples,

表示该线段在时间轴上的长度，r_j表示该线段数值的增长率，即对于线段数据

对于窗口W_j数据线段化后的新序列

得到其三元组序列表示形式为{s₁，s₂，...，s_n}，其中n表示该窗口内数据分段后的线段数量；where k _j represents the slope of the line segment,

Represents the length of the line segment on the time axis, and r _j represents the growth rate of the line segment value, that is, for the line segment data

For the new sequence of window W _j data segmented

Obtain its triplet sequence representation as {s ₁ , s ₂ , ..., s _n }, where n represents the number of line segments after data segmentation in the window;

参数V^B的分段点为

其中n_A和n_B分别表示运行参数A和B的测量数据在该窗口内分段后的线段数；对参数V^A和V^B的分段点进行合并，去除重复项并进行从小到大排列，得到两个参数的分割点序列为

随后，根据该分割点序列以及三元组表示形式获得参数V^A和V^B新的三元组序列为

和

Step 2.3: In the correlation calculation stage, for the two operating parameters ^VA and ^VB during the operation of the equipment, the segmented data needs to be segmented first: take the operating parameter ^VA and the operating parameter ^VB as an example , in the window W _j , the segment point of the parameter V ^A is denoted as

The segmentation point of parameter ^VB is

Among them, n _A and n _B respectively represent the number of line segments of the measurement data of operating parameters ^A and B in this window; the segment points of parameters VA and ^VB are merged, duplicates are removed, and they are arranged from small to large , the split point sequence of the two parameters is obtained as

Then, the parameters VA and ^VB are obtained according to the split point sequence and the triplet representation ^. The new triplet sequence is

and

After obtaining the triplet sequence of the two parameters VA and ^VB in the window ^W _j , the correlation between the two parameters VA and ^VB in the window is calculated by the correlation index d ^AB based on the Euclidean distance ^:

为参数V^A第i条线段的斜率，

为参数V^B第i条线段的斜率，

为参数V^A第i条线段的数值增长率，

为参数V^B第i条线段的数值增长率。where:

is the slope of the ith line segment of parameter V ^A ,

is the slope of the ith line segment of parameter V ^B ,

is the numerical growth rate of the ith line segment of parameter V ^A ,

is the numerical growth rate of the ith line segment of parameter V ^B.

Step 3: construct a relevance prediction model according to the calculated relevance data, the relevance prediction model is a multivariate autoregressive model, and the model parameters are estimated by a partial least squares algorithm. This step specifically includes the following sub-steps:

Step 3.1: Construct the correlation prediction model: set the prediction step size to f, determine the model design parameters U and M according to the known data length so that U+f+M-1=the total number of windows W _j , and construct the following matrix:

Among them, {d ₁ , d ₂ , ..., d _U+f+M-1 ^} represents the correlation sequence of the parameters VA and ^VB , and d ^j represents the two parameters VA and _VB in the window ^W _j Relevance index d ^AB ;

Step 3.2: For each correlation sequence, use the U correlation values to predict the correlation value after f step, and construct a correlation prediction model:

F _p =D _p θ

The parameters θ=[θ ₁ , θ ₂ , . . . , θ _U ] ^T can be obtained by partial least squares algorithm.

Step 4: For the newly obtained data, predict the abnormal working conditions of the equipment according to the correlation prediction model, and continuously update the correlation prediction model and model parameters before the occurrence of abnormal working conditions is predicted. This step specifically includes the following sub-steps:

When predicting, construct a new matrix with the newly acquired data from the sensor:

Subsequently, the correlation value is predicted in f-steps using the constructed correlation prediction model, i.e.

判断设备出现异常，其中d_normal为设备运行初期处于正常运行状态时参数V^A和V^B的关联性值，ω_p为关联性值相对于初始正常的漂移量阈值；若

则利用传感器获得的新数据重构模型以更新模型参数θ；when

It is judged that the equipment is abnormal, where d _normal is the correlation value of the parameters VA and ^VB when the equipment is in ^a normal operating state in the early stage of operation, and ω _p is the drift threshold of the correlation value relative to the initial normal; if

Then use the new data obtained by the sensor to reconstruct the model to update the model parameter θ;

With the update of the data, the correlation value of a certain prediction step is continuously predicted, so as to predict the occurrence time of abnormal working conditions of the equipment. In this example, the prediction step size is f=10, and the threshold _ωp =0.1.

According to steps 2 to 4, correlation calculation and correlation change trend prediction are performed, and the results are shown in Figure 2. Then, according to the set failure threshold ω _p , the abnormal working condition is predicted, and the result is shown in Fig. 3. At the same time, the failure time detected by the correlation sequence obtained by using the actual data according to the set threshold value is also given.

Figure 1 shows the measured values and predicted results of the steam turbine operating parameters. As can be seen from the figure, the use of exponential smoothing prediction can obtain a better prediction effect for a single parameter. Figure 2 shows the comparison between the trend prediction results of the correlation between the steam turbine vacuum A and other parameters and the true value. It can be seen from the figure that a good prediction effect is obtained. Figure 3 shows the prediction results of the occurrence time of the abnormality of the steam turbine. 1-6 of the abscissa in Figure 3 represent the correlation between vacuum A and the other 6 parameters respectively, and the ordinate represents the predicted occurrence time of abnormal working conditions. It can be seen from Fig. 3 that relatively accurate prediction results have been obtained by using several sets of parameter correlation predictions, that is, the occurrence of abnormal conditions can be predicted in advance by a certain step size. In addition, through the analysis of the original data, it is known that the vacuum A starts to indicate abnormality from the 762nd sampling point, and the decline accelerates, and through the analysis of the correlation trend, it can be seen from Figure 3 that the earliest abnormal working condition is predicted to occur. The time is the 590th sampling point (ID6), that is, the occurrence of the exception was caught earlier. More importantly, finding the parameters corresponding to the earliest abnormality through the trend of correlation changes is helpful for judging the location of the abnormality and plays an important role in eliminating the abnormality.

Claims (1)

1. A method for predicting abnormal working conditions in an industrial process based on correlation analysis of operating parameters, characterized in that it comprises the following steps: Step 1: Use the Holt exponential smoothing model to predict the sequence of measurement values collected by each sensor in the industrial process with a fixed step size; Step 2: Construct triples characterized by the changing trend of industrial process operating parameters in the form of sliding windows, represent the measured value sequence in the data window, and calculate any arbitrary value in the window through the correlation index based on Euclidean distance. The correlation of two operating parameters; Step 3: construct a relevance prediction model according to the calculated relevance data, the relevance prediction model is a multiple autoregressive model, and the model parameters are estimated by a partial least squares algorithm; Step 4: For the newly obtained data, predict the abnormal working conditions of the equipment according to the correlation prediction model, and continuously update the correlation prediction model and model parameters before the occurrence of abnormal working conditions is predicted; The step 1 includes the following sub-steps: Step 1.1: For industrial equipment with multiple sensors, record the number of sensors as N. When the equipment is in operation, the operating parameter values that characterize the operating state of the equipment, that is, sensor data, are continuously collected, and each sensor measurement sequence is recorded as

where K is the sequence length,

represents the measurement value of sensor i at the kth sampling time point; Step 1.2: For sensor i, use the Holt exponential smoothing model to predict its measurement sequence, given the measurement

Its smoothed value can be calculated according to the following formula:

in,

express

the smoothed value of ;

is a linear growth factor, representing the trend after smoothing; α and β are smoothing coefficients, both in the range of (0, 1); the initial conditions of the Holt model are as follows:

Step 1.3: After obtaining the smoothed value of the measured data and the linear growth factor, use the result to predict, predict the value

for:

where l represents the prediction step size, and τ is the sampling interval of the device sensor signal; Step 2 contains the following sub-steps: Step 2.1: Running parameter correlation analysis is divided into three stages: time series segmental fitting, triple representation and correlation calculation; in the time series segmental fitting stage, for a data window with a fixed length of L, it is recorded as a window W _j , running parameters X ⁱ , i=1, 2,...,N, the L data in this window are

where j is the starting point of the window, and the corresponding sampling time is t _j ; if it is assumed that m data in one of the time periods in the window

can be fitted by a line segment, for the measured values in

Its corresponding value in the fitted line segment is

Then the calculation formula of the fitting error ERR of the line segment is:

When performing piecewise linearization on the window data starting from j, from

Begin to perform line segment fitting on it; in the process of data segment fitting, follow the steps below, where the set fitting error threshold is ω _E : Step (1): Set the fitting starting point as

The fitting end point is

h=2; where, for the window W _j with j as the starting point, the initial fitting starting point is

Step (2): For the data

Line segment fitting is performed by means of linear regression, thereby obtaining the corresponding line segment data

Calculate its fitting error ERR according to the fitting error ERR calculation formula; Step (3): If ERR≤ω _E , set h=h+1, and repeat step (2); if ERR >ω _E , save the current fitting end point, reset h=2, and return to step ( 1), use the current fitting end point as the new fitting starting point to perform the fitting of the next part of the data; Repeat the above steps (1) to (3) until all the data in the window W _j have been segmented, that is, the piecewise linearized data is obtained

Step 2.2: In the line segment triple representation stage, a line segment s _j is described in the form of the following triples,

where k _j represents the slope of the line segment,

Represents the length of the line segment on the time axis, and r _j represents the growth rate of the line segment value, that is, for the line segment data

For the new sequence of window W _j data segmented

Obtain its triplet sequence representation as {s ₁ , s ₂ , ..., s _n }, where n represents the number of line segments after data segmentation in the window; Step 2.3: In the correlation calculation stage, for the two operating parameters VA and ^VB during the operation of the equipment, the segmented data needs to be segmented first: in the window ^W _j , record the score of the parameter VA ^. segment point is

The segmentation point of parameter ^VB is

Among them, n _A and n _B respectively represent the number of line segments of the measurement data of the operating parameters ^VA and ^VB after segmenting within the window; the segment points of the parameters VA and ^VB are merged, the duplicates are removed, and the ^sequence is from small to Large permutation, the split point sequence of the two parameters is obtained as

Then, the parameters VA and ^VB are obtained according to the split point sequence and the triplet representation ^. The new triplet sequence is

and

After obtaining the triplet sequence of the two parameters VA and ^VB in the window ^W _j , the correlation between the two parameters VA and ^VB in the window is calculated by the correlation index d ^AB based on the Euclidean distance ^:

where:

is the slope of the ith line segment of parameter V ^A ,

is the slope of the ith line segment of parameter V ^B ,

is the numerical growth rate of the ith line segment of parameter V ^A ,

is the numerical growth rate of the i-th line segment of parameter V ^B ; Step 3 contains the following sub-steps: Step 3.1: Construct the correlation prediction model: set the prediction step size to f, determine the model design parameters U and M according to the known data length so that U+f+M-1=the total number of windows W _j , and construct the following matrix:

Among them, {d ₁ , d ₂ , ..., d _U+f+M-1 ^} represents the correlation sequence of the parameters VA and ^VB , and d ^j represents the two parameters VA and _VB in the window ^W _j Relevance index d ^AB ; Step 3.2: For each correlation sequence, use the U correlation values to predict the correlation value after f step, and construct a correlation prediction model: F _p =D _p θ Wherein, the parameter θ=[θ ₁ , θ ₂ , ..., θ _U ] ^T , which can be obtained by partial least squares algorithm; Step 4 contains the following sub-steps: When predicting, construct a new matrix with the newly acquired data from the sensor:

Subsequently, the correlation value is predicted in f-steps using the constructed correlation prediction model, i.e.

when

It is judged that the equipment is abnormal, where d _normal is the correlation value of the parameters VA and ^VB when the equipment is in ^a normal operating state in the early stage of operation, and ω _p is the drift threshold of the correlation value relative to the initial normal; if

Then use the new data obtained by the sensor to reconstruct the model to update the model parameter θ; With the update of the data, the correlation value of a certain prediction step is continuously predicted, so as to predict the occurrence time of abnormal working conditions of the equipment.

Images