Monitoring method and device for quality data of self-correlation mixing process

Abstract

The invention discloses a monitoring method and device for quality data of a self-correlation mixing process, and relates to the technical field of disease monitoring. The method comprises the following steps: obtaining a group of multivariate self-correlation data in a historical controlled state and a group of multivariate self-correlation data in a historical out-of-control state; the method comprises the following steps: respectively calculating a sample mean value and a covariance matrix, carrying out decorrelation on samples through a Cholesky decomposition method, constructing an MEWMA sequence by utilizing standard independent data after decorrelation, training a classifier model of a random forest, designing a control chart method based on the random forest, and monitoring a linear autocorrelation process; on the basis of data decorrelation, according to the random forest theory, a novel control chart is generated, the modeling and monitoring problems of quality data in the self-correlation mixing process are effectively solved, the trained random forest model automatically learns and captures complex and nonlinear modes of data in controlled and out-of-control states, and the quality of the self-correlation mixing process is improved. And effective modeling and monitoring can be carried out no matter the data is in skewed, multi-peak or completely unknown distribution.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of disease monitoring, and particularly relates to a monitoring method and device for self-correlation mixed process quality data, equipment and a medium. BACKGROUND

[0002] Statistical Process Control (SPC) is a process control method that uses statistics and mathematics to analyze, monitor and mine process data; among them, the control chart is the most commonly used and most intuitive and effective SPC tool, and is widely used in processes such as disease monitoring. In the field of disease monitoring, the commonly used scheme of statistical process control is to first collect historical controlled data, select a traditional control chart and determine the control line; then, combined with the real-time collected process data, the statistical quantity is calculated and plotted on the control chart; once the statistical quantity exceeds the control line, an alarm is issued, the process is determined to be out of control, and medical personnel need to analyze the reasons and take timely measures to restore the process to a controlled state; some common control charts include Shewhart charts, Exponentially Weighted Moving Average (EWMA) charts or Cumulative Sum (CUSUM) charts. These traditional control charts usually assume that process data is independent, identically distributed, and follows a normal distribution; however, in some disease (such as breast cancer) records, it is shown that these observations often have temporal correlation and the data distribution deviates from the normal distribution or even an unknown distribution; directly using traditional control charts will have two serious consequences: on the one hand, the controlled average run length ARL is greatly different from the ideal value; on the other hand, due to the influence of data autocorrelation, the monitoring ability of the control chart is greatly reduced after the process is out of control.

[0003] At present, in medical monitoring, the continuously collected physiological parameters or disease indicators may have long-term trends or periodic fluctuations, resulting in sequence correlation between the data; in order to overcome the challenges brought by autocorrelated data, the most common method is to first model the related data using a time series model, and then monitor the residuals using a traditional control chart; commonly used time series models include the first-order vector autoregressive model VAR(1), the first-order vector moving average model VMA(1) and their combination VARMA(1,1) model.

[0004] However, such models first assume that the data follows a normal distribution and remove autocorrelation through model fitting, so that the residual sequence satisfies the independent and identically distributed condition; but when the actual data, i.e. the continuously collected physiological parameters or disease indicators, have serious non-normality (such as skew distribution, multimodal distribution or unknown distribution), the parameter estimation of the time series model will be biased, resulting in independent or normal residuals, and thus the generated disease indicator control chart is difficult to monitor the autocorrelation mixed process quality data of the patient's disease indicator with high accuracy. SUMMARY

[0005] The embodiment of the present application provides a monitoring method and device for self-correlation mixed process quality data, which can solve the problems in the prior art.

[0006] The embodiment of the present application provides a monitoring method for self-correlation mixed process quality data, comprising the following steps: According to the patient historical disease index database, a plurality of sets of self-correlation data of historical controlled state and uncontrolled state of the disease index under the patient time sequence are obtained; The mean value and the covariance matrix of each set of self-correlation data are obtained; based on the mean value and the covariance matrix of each set, a block covariance matrix of an extended vector formed by the current set of self-correlation data and all time points before it is constructed, and a plurality of sets of block covariance matrices are obtained; For each set of block covariance matrix, the block covariance matrix is decomposed into a lower triangular matrix representing the self-correlation structure of the disease index and the product of its transpose by using the Cholesky decomposition method; according to the lower triangular matrix, the controlled and uncontrolled data of the disease index in each set of self-correlation data are respectively processed by recursive decorrelation to generate a set of standard disease index independent data satisfying the independence and normality approximation assumption; The multivariate exponentially weighted moving average sequence MEWMA of each set of standard disease index independent data is obtained, and a random forest model is trained; the MEWMA sequence of real-time online disease index data is input into the trained random forest model to obtain the control chart statistics of the disease index; the control chart statistics are compared with the preset disease index control chart control limit, and the disease index self-correlation mixed process quality data of the patient is monitored based on the comparison size.

[0007] Preferably, the construction of the block covariance matrix comprises: obtaining a set of self-correlation data of historical controlled state of the patient disease index and a set of self-correlation data of historical uncontrolled state , calculating the mean value μ and the covariance matrix {γ(s), 0 ≤ s ≤ b max} respectively, and constructing the covariance matrix of the current set of self-correlation data samples and the previous b samples { X i-b , X i-b+1 ,..., X i} is expressed as: ; ; ; wherein: .

[0008] Preferably, the generation of independent data for the standard disease indicators includes: Using the Cholisky decomposition method to The disease indicators are decomposed, and recursively decorrelated for both controlled and uncontrolled samples, as shown below: ; in: ; ; when i =− n 0 When +1, b =0, otherwise b =min( b +1, b max ), i = i +1, repeat this step until... i >− τ 0, obtain a set of independent data for standard disease indicators that satisfy the assumptions of independence and normality approximation. .

[0009] Preferably, the calculation formula for the multivariate exponentially weighted moving average (MEWMA) series is as follows: ; ; Where: initial value Defined as the mean vector corresponding to the data obtained after standardizing IC samples.

[0010] Preferably, obtaining the preset disease indicator control chart control limits includes: Obtain a set of autocorrelation data for a historical controlled state. The statistics are used to determine the average running chain length of the controlled state, and a desired average running chain length of ARL0=200 is preset as the design standard for the false alarm rate of the control chart. Preset lower bound h l =1, preset upper bound hᵤ =0, as the search interval for initializing the control limits; and set the control limits L =( + h ᵤ ) / 2 as a candidate value for the initial control limit; Based on statistical measures, Monte Carlo simulation is used. If the simulated ARL is less than ARL0, let... =( + hᵤ ) / 2, hᵤ remain unchanged. L =( + hᵤ ) / 2, continue calculating ARL; if the simulated ARL is greater than ARL0, let hᵤ = ( + hᵤ ) / 2, remain unchanged. L =( + hᵤ ) / 2, continue calculating ARL; Repeat the above steps until ARL≈ARL0, at which point... L These are the control limits for the disease indicator control chart.

[0011] Preferably, the quality data of the autocorrelation mixture process for monitoring patients' disease indicators includes: The MEWMA sequence of real-time online disease indicator data is input into the trained random forest model to obtain the control chart statistics, represented as follows: ; p This indicates the probability that the real-time online disease indicator data belongs to a controlled category; like p i <=L If an abnormality occurs during the judgment process, the system will immediately trigger an alarm, prompting the need to find the cause and take corrective measures. like p i >L If the process is under control, monitoring will continue.

[0012] This invention also provides a monitoring device for quality data of autocorrelation mixed processes, comprising: The data modeling module is used to obtain multiple sets of autocorrelation data on the historical controlled and uncontrolled states of the patient's disease indicators in the time series, based on the patient's historical disease indicator database. Obtain the mean and covariance matrix of each group of autocorrelation data; based on the mean and covariance matrix of each group, construct the block covariance matrix of the extended vector formed by the current group's autocorrelation data and the groups of all previous time points, and obtain multiple groups of block covariance matrices; The relevant processing module is used to decompose each block covariance matrix into a lower triangular matrix representing the autocorrelation structure of the disease indicator and its transpose using the Cholisky decomposition method. Based on the lower triangular matrix, the controlled and uncontrolled data of the disease indicator in each group of autocorrelation data are recursively decorrelated to generate a set of standard independent data of the disease indicator that satisfies the assumptions of independence and normality. The monitoring module is used to acquire the multivariate exponential weighted moving average (MEWMA) sequence of independent data for each set of standard disease indicators and train a random forest model. The MEWMA sequence of real-time online disease indicator data is input into the trained random forest model to obtain the control chart statistics of the disease indicators. The control chart statistics are compared with the preset control limits of the disease indicator control chart. Based on the comparison magnitude, the quality data of the patient's autocorrelation mixture process of disease indicators is monitored.

[0013] This invention also provides an electronic device, including a memory and a processor; The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the steps of a monitoring method for quality data of autocorrelation hybrid processes as described above.

[0014] This invention also provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the steps of a monitoring method for quality data of autocorrelation hybrid processes as described above.

[0015] This invention provides a method and apparatus for monitoring quality data of autocorrelation-based hybrid processes. Compared with the prior art, its advantages are as follows: This invention first constructs a block covariance matrix of the extended vector formed by each group of autocorrelation data and all preceding time point groups. This process first captures the dynamic correlation of the autocorrelation data of disease indicators over time. Then, using the Choliski decomposition method, the block covariance matrix is ​​decomposed into the product of a lower triangular matrix representing the autocorrelation structure of the disease indicators and its transpose. Based on the lower triangular matrix, recursive decorrelation processing is performed on the controlled and uncontrolled data in each group of autocorrelation data to transform the original autocorrelation sequence into a set of standard independent disease indicator data that satisfies the assumptions of independence and normality. This process eliminates the autocorrelation of disease indicator data at its source. To mitigate the distorting effect of sex on monitoring statistics, independent data of standard disease indicators are converted into multivariate exponentially weighted moving average (MEWMA) sequences to train a random forest model to obtain control chart statistics for real-time data. This allows the trained random forest model to automatically learn and capture the complex and nonlinear patterns of disease indicator data under controlled and uncontrolled conditions, regardless of whether the data is skewed, multimodal, or completely unknown. This enables effective modeling and monitoring, allowing for high-precision monitoring of the quality data of patients' autocorrelation mixtures of disease indicators using control charts. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall process of a monitoring method for quality data of autocorrelation hybrid processes provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of breast cancer data monitoring and control, provided by an embodiment of the present invention, for a monitoring method of quality data of autocorrelation mixed processes. Detailed Implementation

[0017] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0018] See Figure 1 This invention provides a method for monitoring quality data of autocorrelation mixed processes, mainly addressing the problem that existing methods cannot accurately monitor autocorrelation process data with unknown distributions. Specifically, it improves the monitoring performance of autocorrelation data when traditional control charts have large errors in monitoring autocorrelation data with unknown distributions, enabling earlier and more accurate detection of disease monitoring anomalies. Specific steps include: Step 1: Obtain a set of autocorrelation data for historical controlled states. And a set of autocorrelation data of historical out-of-control states Calculate their mean values ​​respectively. μ The sum and covariance matrix {γ(s), 0 ≤ s ≤ b} max}, and simultaneously construct the current sample and its b previous samples { X i-b , X i-b+1 , ..., X i The covariance matrix of} is expressed as: .

[0019] .

[0020] .

[0021] in: .

[0022] Step 2: Use the Cholisky decomposition method to... The process involves decomposition, recursively decorrelation of both controlled and uncontrolled samples, as follows: .

[0023] in: ; ;when i =− n 0 When +1, b =0; otherwise b =min( b +1, b max ), i = i +1, repeat the above steps until... i >− τ 0.

[0024] Step 3: Calculate the MEWMA value of the standard independent data after removing the correlation, and use it to train the random forest model. The MEWMA formula is as follows: .

[0025] .

[0026] Wherein, initial value Defined as the mean vector corresponding to the data obtained after standardizing IC samples.

[0027] Step 4: Using the historical controlled data from Step 1, calculate its statistic. Let ARL0 = 200. Use the Monte Carlo simulation method to obtain the control limits of the control chart. (1) Let h l =1, hᵤ =0, control limit L =( + hᵤ ) / 2.

[0028] (2) If the simulated ARL is less than ARL0, let = ( + hᵤ ) / 2, hᵤ remain unchanged. L =( + hᵤ ) / 2, continue calculating ARL.

[0029] (3) If the simulated ARL is greater than ARL0, let hᵤ = ( + hᵤ ) / 2, remain unchanged. L =( + hᵤ ) / 2, continue calculating ARL.

[0030] (4) Repeat the above steps until ARL≈ARL0, where L is the control limit of this control chart.

[0031] The determination of the control limits of the control chart first requires giving the average running chain length ARL0 of the controlled system, and then obtaining it using the Monte Carlo simulation method, with 10,000 simulations.

[0032] Step 5: After the online data has undergone the decorrelation process and MEWMA sequence calculation, it is input into the trained random forest model to obtain the control chart statistics, which represent the probability that the sample belongs to the controlled category. .

[0033] like i Real-time monitoring statistics If the process is deemed to have exhibited obvious abnormalities, the cause needs to be identified and corrected; otherwise, the system is considered to be in a stable operating state and should continue to be maintained.

[0034] Among them, the observations collected in steps one and five are both short-term stationary time series, that is, when the collection time of two samples exceeds a certain range, the two samples no longer have serial correlation.

[0035] In a specific implementation, a real breast cancer dataset is used as an example to further describe the invention; the attributes of the breast cancer are as follows: (1) mass thickness (1-10); (2) cell size uniformity (1-10); (3) cell shape uniformity (1-10); (4) edge adhesion (1-10); (5) size of a single epithelial cell (1-10); (6) naked nucleus (1-10); (7) temperate chromatin (1-10); (8) normal nucleoli (1-10); (9) mitosis (1-10), totaling 699 data points. (Refer to...) Figure 1 The implementation steps of this invention are as follows:

[0036] Step 1: Remove data with missing values ​​from the given 699 data points, resulting in 684 data points without missing values. Calculate the mean of the controlled sample and the mean of the uncontrolled sample from these 684 data points. Covariance Matrix And construct the covariance matrix between the current sample and the previous b samples, setting b max The value is 20, and its covariance matrix is ​​expressed as: .

[0037] in, .

[0038] Step 2: Using the following formula, perform decorrelation processing on the 445 controlled data points and 239 out-of-control data points respectively: .

[0039] in: ; .

[0040] Step 3: Calculate the MEWMA value of the dereferenced data, and train a random forest model using the obtained MEWMA sequence values. When training the random forest model, set the number of decision trees to 300, and the number of features selected by the decision trees to be the arithmetic square root of the data dimension, expressed as: .

[0041] .

[0042] In the formula, λ is set to 0.2 and M is set to 15.

[0043] Step 4: Due to the limited amount of data, in order to prevent the historical controlled data for training the random forest model from completely overlapping with the historical controlled data for determining the control limits, the 445 historical controlled data points were expanded to 1000 using a resampling method; given an average running chain length of ARL = 200 under controlled conditions, the control limits were determined using the Monte Carlo simulation method, with 10000 simulation cycles, and the simulation yielded a control limit L of 0.348; Step 5: To prevent the online monitoring data from completely overlapping with the historical out-of-control data, the 239 historical out-of-control data points were resampled to expand the number to 300. These were then used as the objects of online monitoring. After decorrelating them, the MEWMA sequence was obtained and then input into the random forest model to obtain statistics. These statistics were then used in conjunction with the control limits from Step 4 for monitoring.

[0044] This invention is used for real-time monitoring and early warning of anomalies in multivariate time series data of disease clinical indicators, including but not limited to mass thickness, cell size uniformity, cell shape uniformity, edge adhesion, single epithelial cell size, naked nucleus, temperate chromatin, normal nucleoli, and mitosis in breast cancer diagnosis. This invention achieves early detection of disease progression or abnormalities in the data acquisition system by constructing a monitoring system that integrates decorrelation processing, feature engineering, and intelligent classification.

[0045] Based on the calculated statistics and control limits, a control chart for this embodiment was plotted, as follows: Figure 2 As shown in the figure, the monitoring scheme of the present invention detected the out-of-control signal at the fourth point.

[0046] Based on data decorrelation and random forest theory, this invention proposes a novel control chart that effectively solves the modeling and monitoring problems of quality data in autocorrelated mixed processes. The monitoring process of this invention's control chart is visual, and alarms are triggered promptly when anomalies occur. The implementation steps are as follows: 1. Obtain a set of historical controlled autocorrelation data and a set of historical uncontrolled autocorrelation data, calculate the mean and covariance matrix respectively, and construct the covariance matrix of the current sample and the previous b samples; 2. Use Choliski decomposition to decorrelate the controlled and uncontrolled samples respectively; 3. Calculate the MEWMA value of the standardized independent data after decorrelation and use it to train a random forest model; 4. Determine the control limits of the control chart using historical controlled data and Monte Carlo simulation; 5. Calculate the statistics of the online data using the trained random forest model, and monitor the quality data of the autocorrelation mixed process in conjunction with the control limits mentioned above. This invention has a wide range of applications, especially suitable for monitoring process data in industries such as disease surveillance and production.

[0047] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A monitoring method for quality data of a self-correlated mixing process, characterized in that, The method comprises the following steps: According to the patient historical disease index database, obtain the historical controlled state and uncontrolled state of the disease index of the patient in the time sequence, and obtain a plurality of sets of autocorrelation data of the disease index; Obtain the mean and covariance matrix of each set of autocorrelation data; based on the mean and covariance matrix of each set, construct a block covariance matrix of an extended vector formed by the current set of autocorrelation data and all previous time points, and obtain a plurality of sets of block covariance matrices; For each block covariance matrix, the block covariance matrix is decomposed into a lower triangular matrix representing the autocorrelation structure of the disease index and the product of its transpose by using the Cholesky decomposition method; according to the lower triangular matrix, the controlled and uncontrolled data of the disease index in each set of autocorrelation data are respectively processed by recursive decorrelation to generate a set of standard disease index independent data satisfying the independence and normality approximation assumption; Obtain the multivariate exponentially weighted moving average sequence MEWMA of each set of standard disease index independent data, and train a random forest model; input the MEWMA sequence of real-time online disease index data into the trained random forest model to obtain control chart statistics of the disease index; compare the control chart statistics with the preset disease index control chart control limit, and monitor the disease index autocorrelation mixed process quality data of the patient based on the comparison size.

2. The method of claim 1, wherein, The construction of the block covariance matrix comprises: Autocorrelation data of a group of historical controlled states of a patient's disease indicator and a group of historical uncontrolled states , respectively, calculate their mean μ and covariance matrix {γ(s), 0 ≤ s ≤ b max}, while building the covariance matrix of the current group of autocorrelation data samples and its previous b samples{ X i-b , X i-b+1 ,..., X i} is represented as: ； ； ； wherein: .

3. The method of claim 2, wherein the method is a method of monitoring a quality data of a self-correlation mixing process, characterized in that, The generation of the standard disease index independent data comprises: Using the Cholisky decomposition method to The disease indicators are decomposed, and recursively decorrelated for both controlled and uncontrolled samples, as shown below: ； wherein: ; ; When i =− n 0 +1 when, b =0, otherwise b =min( b +1, b max ), i = i +1, repeat the step until i >− τ 0, obtain a set of standard disease indicator independent data that satisfy the independence and normality approximation assumptions.

4. The method of claim 3, wherein the method further comprises: The calculation formula of the multivariate exponentially weighted moving average sequence MEWMA is: ； ； where: initial value is defined as the mean vector corresponding to the data obtained after normalizing the IC samples.

5. The method of claim 4, wherein the method is a method of monitoring a quality of a self-correlation mixing process, characterized by, The acquisition of the preset disease index control chart control limit comprises: Obtaining autocorrelation data of a set of historical controlled states The statistical quantity is preset with an expected controlled state average run length ARL0=200 as the design standard of control chart false alarm rate. preset lower bound h l =1, preset upper bound hᵤ =0, search interval as an initialization control limit; and taking the control limit L ( + hᵤ ) / 2 as an initial control limit candidate value; Based on the statistic, the Monte Carlo simulation method is used for simulation, if the simulation obtained ARL is less than ARL0, let = ( + hᵤ ) / 2, hᵤ remains unchanged, L =( + hᵤ ) / 2, continue to calculate ARL; if the simulation obtained ARL is greater than ARL0, let hᵤ = ( + h ᵤ ) / 2, remains unchanged, L =( + hᵤ ) / 2, continue to calculate ARL; Repeat the above steps until ARL ~ ARL0, at which point L is the control limit for the disease indicator control chart.

6. The method of claim 5, wherein the method further comprises: The monitoring of the disease index autocorrelation mixed process quality data of the patient comprises: Input the MEWMA sequence of real-time online disease index data into the trained random forest model to obtain control chart statistics, which is represented as: ； p represents a probability that the real-time online disease indicator data belongs to a controlled category; If p i <=L If the determination process is abnormal, the system immediately triggers an alarm to prompt the user to find the cause and take corrective measures. If p i >L then the decision process is in a controlled state and continues to remain monitored.

7. A monitoring device for quality data of a self-correlated mixing process, characterized in that It comprises: A data modeling module is configured to obtain a plurality of sets of autocorrelation data of a disease index of a patient in a time sequence according to a patient historical disease index database; Obtain the mean and covariance matrix of each set of autocorrelation data; based on the mean and covariance matrix of each set, construct a block covariance matrix of an extended vector formed by the current set of autocorrelation data and all previous time points, and obtain a plurality of sets of block covariance matrices; A correlation processing module is configured to decompose the block covariance matrix into a lower triangular matrix representing the autocorrelation structure of the disease index and the product of its transpose by using the Cholesky decomposition method for each block covariance matrix; according to the lower triangular matrix, the controlled and uncontrolled data of the disease index in each set of autocorrelation data are respectively processed by recursive decorrelation to generate a set of standard disease index independent data satisfying the independence and normality approximation assumption; The monitoring module is used for obtaining a multivariate exponential weighted moving average sequence MEWMA of each group of standard disease index independent data, training a random forest model, inputting the MEWMA sequence of real-time online disease index data into the trained random forest model, obtaining a control chart statistic of the disease index, comparing the control chart statistic with preset disease index control chart control limits, and monitoring the disease index autocorrelation mixed process quality data of the patient based on the comparison size.

8. An electronic device, comprising: Comprise: a memory and a processor; the memory is used for storing a computer program; the processor is used for executing the computer program stored in the memory to realize the steps of the monitoring method for the autocorrelation mixed process quality data according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, a computer program is stored, and the computer program is executed by the processor to realize the steps of the monitoring method for the autocorrelation mixed process quality data according to any one of claims 1-6.

Images