Abnormal state automatic identification and correction system of radiopharmaceutical production line

Abstract

The invention discloses an abnormal state automatic identification and correction system for a radiopharmaceutical production line, and relates to the technical field of pharmaceutical production. The data collection module is used for collecting production line data and a data standard interval, and distinguishing an abnormal data set and a pseudo normal data set; the data processing module is used for judging the number of associated samples of the abnormal sample data; obtaining a net normal data set; establishing data type pairs, calculating correlation factors of the data type pairs, performing grade division and grouping on the data type pairs, and calculating a difference threshold value of each group of data; and calculating a predicted abnormal value. According to the method, by associating the sample number, obtaining the net normal data set, eliminating the influence of data fluctuation before abnormality, and comparing the difference index of the new data with the difference threshold value, when the abnormality index of the new data exceeds a certain range, it is judged that the production line data is likely to be abnormal, the prediction effect is achieved, and correction is conducted in time; and possible abnormal conditions are pre-judged, so that the occurrence of abnormal states is reduced.

Description

Technical Field

[0001] This invention relates to the field of pharmaceutical manufacturing technology, specifically to an automatic identification and correction system for abnormal conditions in a radiopharmaceutical production line. Background Technology

[0002] In the field of radiopharmaceutical production, production lines operate under complex conditions with extremely high safety and quality requirements. Failure to identify abnormal conditions in a timely manner can lead to serious consequences such as increased risk of radioactive material leakage and substandard drug quality. Automatic abnormal condition identification technology can monitor various parameters and operating statuses of the production line in real time, quickly and accurately capture abnormal signals, issue timely alarms, and assist in taking countermeasures. This not only helps ensure production safety and improve drug quality stability but also increases production efficiency and reduces potential losses.

[0003] The accuracy of anomaly detection in radiopharmaceutical production lines largely depends on the methods used. For example, a digital twin-based method and system for processing anomaly data in a production line, as disclosed in patent publication number CN119105422A, can collect digital twin model data of the production line to determine the number of devices and the data analysis period. It collects the operating data of each device within the analysis period, arranges it into a data matrix in chronological order, and determines whether any devices exhibit anomalies. When an anomaly is detected, it extracts the anomaly features and compares them with historical operating data to determine the fault type. Based on the fault type and anomaly features, it determines the risk value of the devices and ranks the devices with anomalies. Finally, it iterates through the historical operating data of each anomalous device to obtain historical impact values ​​and determine whether the ranking needs adjustment, resulting in the final ranking of anomalous devices. This improves the accuracy of anomaly detection in the production line and reduces its maintenance costs.

[0004] Radiopharmaceutical production lines typically operate for extended periods. To ensure production quality, it is necessary to monitor production line data and promptly correct any abnormalities. Some abnormalities occur as production time progresses and gradually deviates from standard values. While the methods described above and those based on the same principle can monitor abnormalities, they cannot predict potential anomalies based on changes in production line data before they occur, thus reducing the likelihood of anomalies. Consequently, the effectiveness of anomaly detection is relatively poor. Summary of the Invention

[0005] The purpose of this invention is to provide an automatic identification and correction system for abnormal states in a radiopharmaceutical production line, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an automatic identification and correction system for abnormal states in a radiopharmaceutical production line, comprising: Data collection module: Collects production line data and standard data ranges, establishes a total database, and classifies and extracts sample data under abnormal conditions and sample data under normal conditions to obtain abnormal datasets and pseudo-normal datasets. Data processing module: Each abnormal sample data is marked as an anomaly. The normal sample data preceding each abnormal sample data is checked in reverse chronological order. The number of associated samples of the abnormal sample data is determined based on the fluctuation prediction method. Based on the number of associated samples for each abnormal sample data, remove the corresponding normal sample data from the pseudo-normal dataset to obtain the net normal dataset. In the net normal dataset, data type pairs are established between every two data types. The correlation factor of the data type pairs is calculated according to the correlation factor calculation method, and the data type pairs are classified into levels according to the magnitude of the correlation factor. Group the data types in each level, calculate the difference index for each group, and select the largest difference index as the difference threshold. The latest production line data is compared with the standard data range to identify a number of outlier-like points. For each type of outlier, the difference index of each group of data with the target type of outlier is compared with the difference threshold, and the sum is obtained after weighting according to different classification levels to obtain the predicted outlier value. Output module: Compares the maximum value of the predicted outlier with the preset threshold. When the predicted outlier is greater than the preset threshold, it determines that the data may be abnormal and performs correction according to the preset correction process.

[0007] Preferably, the fluctuation prediction method includes: Select an abnormal sample data, and extract the data with the same data type as the abnormal parameter item from the first few normal sample data based on its abnormal data type item; Select the first few samples of the target anomaly data, calculate the mean and standard deviation, and then calculate the number of associated samples according to the formula, which is as follows: ; Where X t This represents outlier sample data at time point t. This represents the mean of the first k normal samples from the target abnormal sample data. The standard deviation of the first k normal samples from the target abnormal sample data is represented by s, where s represents the number of associated samples. Indicates related indicators.

[0008] Preferably, the fluctuation prediction method includes: Select an abnormal sample data and extract the first few normal sample data; For each data type item, calculate its global mean and standard deviation; Calculate the standard matrix B of the sample data at time point t using the formula. t The formula is as follows: ; Where i represents the data type item, μ i σ represents the mean of all data in data type item i. i b represents the standard deviation of all data in data type item i. ti The standard matrix B representing the sample data t The i-th eigenvalue, X ti This represents the value of data type item i in the sample data at time point t; The number of associated samples is calculated using the following formula: ; Where i represents a data type item, d represents the total number of data type items, and b ti B represents the standard matrix of outlier sample data at time point t. t The i-th eigenvalue, b mi B represents the standard matrix of normal sample data at time point m. t The i-th eigenvalue, The eigenvalue of the i-th eigenvalue in the standard matrix representing the first k normal samples of the target abnormal sample data is denoted as s, where s represents the number of associated samples. Indicates related indicators.

[0009] Preferably, the method for calculating the correlation factor includes: Randomly select two data type items; Sort the data items of the two data types separately, obtain their ranks, and calculate the difference between the ranks; The correlation factor is calculated based on the absolute value of the Spearman rank correlation coefficient.

[0010] Preferably, the data type pairs within each level are grouped. Specifically, this includes: For data type pairs of the same level, group two data type pairs with the same data type together, and continue to group other data type pairs with the same data type into the same group. For data type pairs that do not have the same data type, group them separately.

[0011] Preferably, the calculation of the difference threshold for each group of data specifically includes: Using the net normal dataset as the data source, select the data type item corresponding to each group of data, and calculate the mean and standard deviation of each data type; The normal sample data in the net normal dataset is transformed according to the formula to obtain standard data. The specific formula is as follows: ; Where f ti x represents the standard data of the i-th data type item in the sample at time point t. ti μ represents the original value of the i-th data type item in the sample at time point t. i σ represents the mean of all data in data type item i. i This represents the standard deviation of all data in data type item i; Then, the difference index for each standard data point is calculated, specifically: ; in f represents the difference index of samples at time point t. ti Let d represent the standard data of the i-th data type item in the sample at time point t, and d represent the total number of data type items in the sample. Select the largest difference indicator as the difference threshold for the target group.

[0012] Preferably, the step of comparing the latest acquired production line data with the data standard range to obtain a number of outlier-like points specifically includes: Select one of the data types from the latest acquired production line data and calculate the proportion of the target data value within the corresponding data standard range; Sort the position proportions in order of size, and select the top and bottom data type items as class anomalies.

[0013] Preferably, obtaining the predicted outliers specifically includes: Different weights are assigned to different levels, specifically: ; in represents the weight of level c, and g represents the total number of levels; Calculate the ratio of the difference index of different groups in each level to the difference threshold of the target group, and calculate the mean; The mean is weighted by the grade weights and then summed to obtain the predicted outliers.

[0014] Compared with the prior art, the beneficial effects of the present invention are: By correlating sample numbers, pseudo-normal datasets that may have appeared before anomalies are removed, resulting in a net normal dataset. This eliminates the impact of data fluctuations before anomalies occur. Data types are then grouped according to relevant factors, and a difference threshold is calculated for each group. The difference indicators of new data are compared with the difference thresholds. When the abnormal indicators of new data exceed a certain range, it is determined that production line data may be abnormal, achieving a predictive effect and timely correction. This allows for the prediction of possible abnormal situations, thereby reducing the occurrence of abnormal states.

[0015] Meanwhile, when comparing new data with historical data, the new data first identifies several points of difference and calculates the ratio of the difference index to the difference threshold. This eliminates the need to calculate for each data type separately, reducing computational load while ensuring prediction accuracy.

[0016] Furthermore, when classifying and grouping data based on relevant factors, it is possible to distinguish data pairs with different degrees of correlation. When calculating the predicted outliers of new data, different weighting coefficients can be set at different levels, which can further improve the accuracy of predicting outliers. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the automatic identification and correction of abnormal states according to the present invention. Figure 2 This is a flowchart illustrating the fluctuation prediction method in this invention; Figure 3 This is a schematic diagram illustrating the calculation process of the difference threshold in this invention; Figure 4 This is a schematic diagram of the calculation process for predicting outliers in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In this application, for ease of understanding, the method steps used do not necessarily need to be executed in the order of steps in this embodiment during actual operation. In other embodiments, these steps may be performed simultaneously or in a different order.

[0020] Example 1:

[0021] Radiopharmaceutical production lines typically operate for extended periods. To ensure production quality, it is necessary to monitor production line data and anticipate potential anomalies based on changes in the production line data before they occur, thereby reducing the likelihood of abnormal situations.

[0022] like Figures 1-4 As shown, the present invention provides a technical solution: an automatic identification and correction system for abnormal states in a radiopharmaceutical production line, comprising: Data collection module: Collects production line data and standard data ranges, establishes a total database, and classifies and extracts sample data under abnormal conditions and sample data under normal conditions to obtain abnormal datasets and pseudo-normal datasets. Data processing module: Each abnormal sample data is marked as an anomaly. The normal sample data preceding each abnormal sample data is checked in reverse chronological order. The number of associated samples of the abnormal sample data is determined based on the fluctuation prediction method. Based on the number of associated samples for each abnormal sample data, remove the corresponding normal sample data from the pseudo-normal dataset to obtain the net normal dataset. In the net normal dataset, data type pairs are established between every two data types. The correlation factor of the data type pairs is calculated according to the correlation factor calculation method, and the data type pairs are classified into levels according to the magnitude of the correlation factor. Group the data types in each level, calculate the difference index for each group, and select the largest difference index as the difference threshold. The latest production line data is compared with the standard data range to identify a number of outlier-like points. For each type of outlier, the difference index of each group of data with the target type of outlier is compared with the difference threshold, and the sum is obtained after weighting according to different classification levels to obtain the predicted outlier value. Output module: Compares the maximum value of the predicted outlier with the preset threshold. When the predicted outlier is greater than the preset threshold, it determines that the data may be abnormal and performs correction according to the preset correction process.

[0023] It is important to note that production line data can be acquired through sensors, including but not limited to reactor temperature, pressure, pH value, and ambient radiation dose rate. Moreover, technical personnel set standard ranges for these data types. For example, the standard range for ambient radiation dose rate can be set to 1.3μSv / h-4.8μSv / h to ensure environmental safety; the reactor temperature can be set to 24℃-26℃ to ensure the stability of the chemical reaction. Other data types also need to have standard ranges set by technical personnel according to actual needs.

[0024] An abnormal state is defined as a situation where, at a certain detection point in time, the value of any data type exceeds its standard range.

[0025] like Figure 2 As shown, the methods for predicting fluctuations include: Select an abnormal sample data, and extract the data with the same data type as the abnormal parameter item from the first few normal sample data based on its abnormal data type item; Select the first few samples of the target anomaly data, calculate the mean and standard deviation, and then calculate the number of associated samples according to the formula, which is as follows: ; Where X t This represents outlier sample data at time point t. This represents the mean of the first k normal samples from the target abnormal sample data. The standard deviation of the first k normal samples from the target abnormal sample data is represented by s, where s represents the number of associated samples. Indicates related indicators.

[0026] It should be noted that, for ease of understanding, the following simulated data is used: Assuming that the standard range for the ambient radiation dose rate in the production line data can be set to 1.3 μSv / h-4.8 μSv / h; The samples from the four tests are shown in Table 1 below: Table 1: Detection Sample Table (Single Element) Time point Ambient radiation dose rate Anomaly Detection 1 2.3 normal 2 2.8 normal 3 3.4 normal 4 5.2 abnormal The abnormal sample data is the sample data at time point 4 (the first three times are recorded in the pseudo-normal dataset, and the fourth time is recorded in the abnormal dataset. For ease of understanding, they are shown in a table here). Therefore, t=4, so the following calculations are performed. For k=1: t=4, t−k=3, t−1=3; μ 3:3 =x3=3.4; σ 3:3 =0 (because there is only one data point x3, and the standard deviation is 0).

[0027] If the denominator is 0, calculate and skip.

[0028] For k=2: t=4, t−k=2, t−1=3; μ 3:3 =0.5×(x3+x2)=2.85; σ 3:3 ≈0.55.

[0029] ≈4.27>3 Therefore, s = 2 (k = 3 and subsequent values ​​can be disregarded).

[0030] After calculating the number of associated samples, the net normal dataset is obtained by removing the first few normal samples (the number of which is equal to the number of associated samples corresponding to the abnormal sample) at the time point corresponding to each abnormal sample data in the pseudo normal dataset.

[0031] The methods for calculating correlation factors include: Randomly select two data type items; Sort the data items of the two data types separately, obtain their ranks, and calculate the difference between the ranks; The correlation factor is calculated based on the absolute value of the Spearman rank correlation coefficient.

[0032] It should be noted that, for ease of calculation, the following simulated data is used: Assuming the production line data includes: reactor temperature (24℃-26℃ during standard period), reactor pressure (1.0 bar-1.5 bar during standard period), ambient radiation dose rate (1.3 μSv / h-4.8 μSv / h during standard period), and pH value (7.0-7.4 during standard period), the following is the production line data (net normal dataset) for the past 5 periods, as shown in Table 2: Table 2: Demonstration Data of Relevant Factors sample Reactor temperature Reactor pressure Ambient radiation dose rate pH value 1 24.4 1.1 2.4 7.2 2 24.6 1.3 2.6 7.1 3 25.0 1.2 3.2 7.2 4 24.8 1.3 2.5 7.3 5 25.1 1.1 2.1 7.2 Taking reactor temperature and ambient radiation dose rate as examples, the relevant factors are calculated as follows: First, the data on reactor temperature and ambient radiation dose rate were sorted as (5, 4, 2, 3, 1) and (4, 2, 1, 3, 5), respectively. Then, the Spearman rank correlation coefficient was calculated using the following formula: ; in This represents the correlation factor between the Xth data type and the Yth data type, where n represents the total number of samples, and d... i 2 (X, Y) represents the square of the difference between the rank of the Xth data type and the rank of the Yth data type in sample i.

[0033] Calculate d i 2 (X, Y) are (1, 4, 1, 0, 16), and the formula is used to calculate... =-0.1, therefore the correlation factor is 0.1. The calculation method for the other elements is the same, and will not be elaborated here.

[0034] The closer the correlation factor is to 0, the weaker the correlation between the two data types, indicating that the prediction of abnormal states through the data fluctuations between these two data types is less effective. Conversely, the closer the correlation factor is to 1, the stronger the correlation between the two data types, indicating that the prediction of abnormal states through the data fluctuations between these two data types is more accurate.

[0035] When classifying data types into levels, several level factors can be set (the more level factors, the more levels, the greater the computational load, but the more accurate the prediction results; the classification can be made according to actual needs, and there are no restrictions on this). For example, if the level factors are 0.2 and 0.5, the correlation factor less than 0.2 is assigned to the first level, the correlation factor greater than or equal to 0.2 and less than 0.5 is assigned to the second level, and the correlation factor greater than or equal to 0.5 is assigned to the third level.

[0036] Grouping data types within each level, specifically including: For data type pairs of the same level, group two data type pairs with the same data type together, and continue to group other data type pairs with the same data type into the same group. For data type pairs that do not have the same data type, group them separately.

[0037] It should be noted that, for ease of understanding, the following simulated data is used: Assume that in the second level, there are 4 data type pairs: reactor temperature and reactor pressure, reactor pressure and ambient radiation dose rate, ambient radiation dose rate and source surface radiation dose rate, stirring speed and pH value. Among them, the two pairs of reactor temperature and reactor pressure, and reactor pressure and ambient radiation dose rate have the same data type of reactor pressure. The two pairs of reactor pressure and ambient radiation dose rate, and ambient radiation dose rate and source surface radiation dose rate have the same data type of ambient radiation dose rate. Therefore, these three pairs are combined into one group, and the data type in this group is (reactor temperature, reactor pressure, ambient radiation dose rate, source surface radiation dose rate). The data type pair of stirring speed and pH value does not share the same data type as other data type pairs, so this pair is grouped separately, and the data type in this group is (stirring speed, pH value).

[0038] like Figure 3 As shown, the difference threshold for each group of data is calculated, specifically including: Using the net normal dataset as the data source, select the data type item corresponding to each group of data, and calculate the mean and standard deviation of each data type; The normal sample data in the net normal dataset is transformed according to the formula to obtain standard data. The specific formula is as follows: ; Where f ti x represents the standard data of the i-th data type item in the sample at time point t. ti μ represents the original value of the i-th data type item in the sample at time point t. i σ represents the mean of all data in data type item i. i This represents the standard deviation of all data in data type item i; Then, the difference index for each standard data point is calculated, specifically: ; in f represents the difference index of samples at time point t. ti Let d represent the standard data of the i-th data type item in the sample at time point t, and d represent the total number of data type items in the sample. Select the largest difference indicator as the difference threshold for the target group.

[0039] It should be noted that, for ease of understanding, the following simulated data is used: Suppose we have a set of data as shown in Table 3 below: Table 3: Demonstration Data for Difference Thresholds Time point Ambient radiation dose rate μSv / h Source surface radiation dose rate μSv / h 1 2.4 120 2 2.6 123 3 2.8 124 First, calculate the mean and standard deviation: μ 1 =2.6, μ 2 ≈0.2, σ 1 =122.33, σ 2 ≈1.53.

[0040] For time point 1: f 11 =-1, f 12 =-1.53, therefore ≈1.83; For time point 2: f 21 =0, f 22 ≈0.44, therefore ≈0.44; For time point 3: f 31 =1, f 32 ≈1.09, therefore ≈1.48.

[0041] The largest difference metric was chosen as 1.83, therefore the difference threshold was 1.83.

[0042] The latest production line data is compared with the standard data range to identify a number of outlier-like points, specifically including: Select one of the data types from the latest acquired production line data and calculate the proportion of the target data value within the corresponding data standard range; Sort the position proportions in order of size, and select the top and bottom data type items as class anomalies.

[0043] It should be noted that, for ease of understanding, the following simulated data is used: Assume the latest monitored ambient radiation dose rate is 2.9, while the standard period for ambient radiation dose rate is 1.3 μSv / h-4.8 μSv / h.

[0044] First, calculate the location ratio (the location of the monitored data within the standard range): (2.9-1.3) ÷ (4.8-1.3) ≈ 0.46.

[0045] Assuming there are four other data types with position ratios of 0.52, 0.68, 0.23, and 0.36 respectively, select the two highest and two lowest data type items from the five position ratios as the class anomalies (here, four data type items are selected in total, the highest and lowest; in actual use, there are more data type items, more options, and more data type items selected, the higher the accuracy of the calculation, but the larger the amount of calculation. The selection can be made according to actual needs, and there is no specific restriction).

[0046] like Figure 4 As shown, obtaining predicted outliers specifically includes: Different weights are assigned to different levels, specifically: ; in represents the weight of level c, and g represents the total number of levels; Calculate the ratio of the difference index of different groups in each level to the difference threshold of the target group, and calculate the mean; The mean is weighted by the grade weights and then summed to obtain the predicted outliers.

[0047] It should be noted that, for ease of understanding, the following simulated data is used: Assume there are three different levels. The third level has three groups with difference indices of 1.6, 1.5, and 1.4 calculated from the new production line data, and corresponding difference thresholds of 1.9, 1.6, and 2.0, respectively. The second level has one group with a difference index of 0.9 calculated from the new production line data, and a corresponding difference threshold of 1.5. The first level has one group with a difference index of 2.3 calculated from the new production line data, and a corresponding difference threshold of 2.4. First, calculate the weight of each level according to the formula. First level: ≈0.14, ≈0.28, ≈0.57 (Rounding is used here to ensure the sum is not 1; theoretically, the sum of the weights of the three levels equals 1). Calculate the ratio of the difference index to the difference threshold of the target group, and calculate the mean: For the third level: the proportions of the three groups are 1.6÷1.9≈0.84, 1.5÷1.6≈0.94, and 1.4÷2.0=0.7, with a calculated mean of approximately 0.82; For the second level: the proportion is 0.9 ÷ 1.5 = 0.6, and the mean is also 0.6; For the first level: the ratio is 2.3 ÷ 2.4 = 0.96, and the mean is also 0.96; The weighted summation based on the rank weights is 0.7698, therefore the predicted outlier is 0.7698.

[0048] Finally, by comparing the predicted outliers with the preset threshold, it can be determined whether the production line is about to experience an abnormal state. Here, we assume that the preset threshold is 0.8, then the prediction result is that no abnormal state will occur. The preset threshold needs to be determined according to the actual situation and can be between 0 and 1. The smaller the value, the higher the prediction sensitivity, but the easier it is to falsely report anomalies. The larger the value, the lower the prediction sensitivity, but the more accurate the prediction result. There is no limit to the specific value.

[0049] Furthermore, when setting up the correction process, different correction processes can be set according to different anomalies. For example, when the temperature is high and abnormal, the power of the refrigeration equipment can be increased. When the prediction result is abnormal, the anomaly points involved in this prediction are retrieved in reverse. First, a anomaly point of the highest level is randomly selected. The correction process of this anomaly point is fine-tuned and the feedback data is detected. If the production line data recovers (no longer predicts anomalies), the correction process is no longer performed. If the production line data still does not recover, the correction process corresponding to the next anomaly point is adjusted. The anomaly points of the same level are randomly selected for adjustment from high to low level until the production line data no longer predicts anomalies. If the correction processes corresponding to all anomaly points cannot restore the production line data, the technicians are notified to adjust it manually.

[0050] Example 2:

[0051] In Example 1, when calculating the number of associated samples, only the abnormal data type item is considered. The calculation of a single data item is simple, but other data type items may show abnormal signs earlier. Therefore, the determination of the number of associated samples is not accurate enough. Based on this, this example provides another fluctuation prediction method to calculate the number of associated samples.

[0052] Fluctuation prediction methods include: Select an abnormal sample data and extract the first few normal sample data; For each data type item, calculate its global mean and standard deviation; Calculate the standard matrix B of the sample data at time point t using the formula. t The formula is as follows: ; Where i represents the data type item, μ i σ represents the mean of all data in data type item i. i b represents the standard deviation of all data in data type item i. ti The standard matrix B representing the sample data t The i-th eigenvalue, X ti This represents the value of data type item i in the sample data at time point t; The number of associated samples is calculated using the following formula: ; Where i represents a data type item, d represents the total number of data type items, and b ti B represents the standard matrix of outlier sample data at time point t. t The i-th eigenvalue, b mi B represents the standard matrix of normal sample data at time point m. t The i-th eigenvalue, The eigenvalue of the i-th eigenvalue in the standard matrix representing the first k normal samples of the target abnormal sample data is denoted as s, where s represents the number of associated samples. Indicates related indicators.

[0053] It should be noted that, for ease of understanding, the following simulated data is used: Assuming the production line data includes two data types, temperature and humidity (d=2), the seven monitoring samples are shown in Table 4 below: Table 4: Detection Sample Table (Multi-element) Time point temperature humidity Anomaly Detection 1 20 10 normal 2 22 12 normal 3 21 11 normal 4 23 13 normal 5 50 30 abnormal 6 24 14 normal 7 25 15 normal The outlier sample data is the sample data at time point 5 (the fifth record is in the outlier dataset, and is shown in a table here for ease of understanding). The following calculations are performed: Calculate the mean μ 1 ≈26.43, μ 2 =15, σ 1 ≈9.75, σ 2 ≈6.32; For t=5 (outlier sample data): b 51 ≈2.42, b 52 ≈2.37; When k=1, calculate t=5 and t=4: When t=5, the above calculation has already been performed; When t=4, b 41 ≈-0.35, b 42 ≈-0.32; ≈3.86<4, continue calculation; When k=2, calculate t=5, t=4, and t=3: The above calculations have already been performed for t=5 and t=4; When t=3, b 41 ≈-0.56, b 42 ≈-0.63; ≈4.04>4, therefore the number of associated samples s=2.

[0054] Taking into account the impact of multiple data types on outliers, when other parameters cause anomalies, the abnormal signs of other parameters appear earlier. Therefore, compared with Example 1, this example takes into account the data fluctuations of multiple data types and calculates the time of anomaly signs (number of associated samples) more accurately.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended embodiments and their equivalents.

Claims

1. An automatic identification and correction system for abnormal conditions in a radiopharmaceutical production line, comprising: Data collection module: Collects production line data and standard data ranges, establishes a total database, and classifies and extracts sample data under abnormal conditions and sample data under normal conditions to obtain abnormal datasets and pseudo-normal datasets. Its features are: Data processing module: Each abnormal sample data is marked as an anomaly. The normal sample data preceding each abnormal sample data is checked in reverse chronological order. The number of associated samples of the abnormal sample data is determined based on the fluctuation prediction method. Based on the number of associated samples for each abnormal sample data, remove the corresponding normal sample data from the pseudo-normal dataset to obtain the net normal dataset. In the net normal dataset, data type pairs are established between every two data types. The correlation factor of the data type pairs is calculated according to the correlation factor calculation method, and the data type pairs are classified into levels according to the magnitude of the correlation factor. Group the data types in each level, calculate the difference index for each group, and select the largest difference index as the difference threshold. The latest production line data is compared with the standard data range to identify a number of outlier-like points. For each type of outlier, the difference index of each group of data with the target type of outlier is compared with the difference threshold, and the sum is obtained after weighting according to different classification levels to obtain the predicted outlier value. Output module: Compares the maximum value of the predicted outlier with the preset threshold. When the predicted outlier is greater than the preset threshold, it determines that the data may be abnormal and performs correction according to the preset correction process.

2. The automatic identification and correction system for abnormal states in a radiopharmaceutical production line according to claim 1, characterized in that: The fluctuation prediction method includes: Select any abnormal sample data, and extract the data with the same data type as the abnormal parameter item from the first few normal sample data based on its abnormal data type item; Select the first few samples of the target anomaly data, calculate the mean and standard deviation, and then calculate the number of associated samples according to the formula, which is as follows: ； Where X t This represents outlier sample data at time point t. This represents the mean of the first k normal samples from the target abnormal sample data. The standard deviation of the first k normal samples from the target abnormal sample data is represented by s, where s represents the number of associated samples. Indicates related indicators.

3. The automatic identification and correction system for abnormal states in a radiopharmaceutical production line according to claim 1, characterized in that: The fluctuation prediction method includes: Select an abnormal sample data and extract the first few normal sample data; For each data type item, calculate its global mean and standard deviation; Calculate the standard matrix B of the sample data at time point t using the formula. t The formula is as follows: ； Where i represents the data type item, μ i σ represents the mean of all data in data type item i. i b represents the standard deviation of all data in data type item i. ti The standard matrix B representing the sample data t The i-th eigenvalue, X ti This represents the value of data type item i in the sample data at time point t; The number of associated samples is calculated using the following formula: ； Where i represents a data type item, d represents the total number of data type items, and b ti B represents the standard matrix of outlier sample data at time point t. t The i-th eigenvalue, b mi B represents the standard matrix of normal sample data at time point m. t The i-th eigenvalue, The eigenvalue of the i-th eigenvalue in the standard matrix representing the first k normal samples of the target abnormal sample data is denoted as s, where s represents the number of associated samples. Indicates related indicators.

4. The automatic identification and correction system for abnormal states in a radiopharmaceutical production line according to claim 1, characterized in that: The method for calculating the relevant factors includes: Randomly select two data type items; Sort the data items of the two data types separately, obtain their ranks, and calculate the difference between the ranks; The correlation factor is calculated based on the absolute value of the Spearman rank correlation coefficient.

5. The automatic identification and correction system for abnormal states in a radiopharmaceutical production line according to claim 1, characterized in that: The grouping of data type pairs within each level specifically includes: For data type pairs of the same level, group two data type pairs with the same data type together, and continue to group other data type pairs with the same data type into the same group. For data type pairs that do not have the same data type, group them separately.

6. The automatic identification and correction system for abnormal states in a radiopharmaceutical production line according to claim 1, characterized in that: The calculation of the difference threshold for each group of data specifically includes: Using the net normal dataset as the data source, select the data type item corresponding to each group of data, and calculate the mean and standard deviation of each data type; The normal sample data in the net normal dataset is transformed according to the formula to obtain standard data. The specific formula is as follows: ； Where f ti x represents the standard data of the i-th data type item in the sample at time point t. ti μ represents the original value of the i-th data type item in the sample at time point t. i σ represents the mean of all data in data type item i. i This represents the standard deviation of all data in data type item i; Then, the difference index for each standard data point is calculated, specifically: ； in f represents the difference index of samples at time point t. ti Let d represent the standard data of the i-th data type item in the sample at time point t, and d represent the total number of data type items in the sample. Select the largest difference indicator as the difference threshold for the target group.

7. The automatic identification and correction system for abnormal states in a radiopharmaceutical production line according to claim 1, characterized in that: The process involves comparing the latest acquired production line data with the standard data range to identify a number of outlier-like points, specifically including: Select one of the data types from the latest acquired production line data and calculate the proportion of the target data value within the corresponding data standard range; Sort the position proportions in order of size, and select the top and bottom data type items as class anomalies.

8. The automatic identification and correction system for abnormal states in a radiopharmaceutical production line according to claim 1, characterized in that: The specific steps of obtaining predicted outliers include: Different weights are assigned to different levels, specifically: ； in represents the weight of level c, and g represents the total number of levels; Calculate the ratio of the difference index of different groups in each level to the difference threshold of the target group, and calculate the mean; The mean is weighted by the grade weights and then summed to obtain the predicted outliers.

Images