A Smart Production Monitoring Method and System for Injectable Solutions
By using the method of local aggregation degree difference and dynamic merging distance threshold, the problem of false alarms and missed alarms caused by fixed thresholds in injection production is solved, and the accurate identification of phase transitions and anomalies is achieved, thereby improving the monitoring accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- RUICHENG LVMAN BIOLOGICAL PHARM CO LTD
- Filing Date
- 2026-04-27
- Publication Date
- 2026-05-26
AI Technical Summary
Existing methods for monitoring the production of injectable solutions rely on fixed thresholds, which can lead to false alarms or missed alarms. Traditional density clustering algorithms cannot effectively distinguish between normal phase transitions and real process anomalies, thus affecting the accuracy of monitoring.
By employing a method that combines local clustering differences and dynamic merging distance thresholds, stage boundary points are identified through the calculation of local clustering differences. The merging distance threshold is dynamically adjusted to allow normal stage boundary points to merge with high-density clusters, thereby reducing the false alarm rate.
It significantly improved the monitoring accuracy during the production process of injection solutions, reduced the false alarm rate, and enhanced the ability to identify phase transitions and anomalies.
Smart Images
Figure CN122087489A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent monitoring technology. More specifically, this invention relates to an intelligent production monitoring method and system for injectable solutions. Background Technology
[0002] In pharmaceutical manufacturing, the production of injection solutions is a critical step. The core process of preparing the drug solution involves a series of chemical changes, such as the dissolution of raw materials and acid-base neutralization. To ensure the quality and safety of the final drug, the state inside the preparation tank must be closely monitored. This can be achieved by monitoring key process parameters such as temperature, pH value, and stirring motor current. These parameters are coupled together to form a dynamic, multi-dimensional time series data reflecting the drug solution preparation process.
[0003] Currently, the main challenge in ensuring the stability of the production process lies in how to accurately identify abnormal operating conditions from these complex multi-dimensional time-series data so that staff can adjust and optimize parameters in a timely manner. Existing monitoring methods mostly rely on setting fixed thresholds for individual parameters and triggering alarms when parameters exceed the limits. However, the drug preparation process has multiple process stages, and the normal parameter distribution characteristics of each stage are significantly different. Fixed thresholds cannot adapt to the dynamic stage changes of this process, which can easily lead to false alarms or missed alarms.
[0004] To address the limitations of fixed threshold methods, density clustering algorithms can be used for anomaly monitoring, considering that outlier data points are usually located in low-density areas. However, when analyzing time-series data, these algorithms ignore the continuity and contextual relationships of data points over time. For example, the stage boundary data generated when drug preparation data transitions from one stable stage to another may also appear as low-density points in the data space. This makes it difficult for the algorithm to effectively distinguish between normal stage transitions and real process anomalies, resulting in low monitoring accuracy and affecting the quality of subsequent drug production. Summary of the Invention
[0005] To address the aforementioned technical problem of difficulty in effectively distinguishing between normal phase transitions and actual process anomalies in actual monitoring, resulting in low monitoring accuracy, this invention provides solutions in the following aspects.
[0006] In a first aspect, the present invention provides an intelligent production monitoring method for injectable solutions, comprising: collecting multi-dimensional data from the injectable solution preparation tank at various times as data points; for any data point, determining the local clustering degree based on the changes in each dimension of its neighboring data points; calculating the difference in the local clustering degree of its temporally adjacent data points as the local clustering degree difference; in response to the local clustering degree difference being greater than a difference threshold, iteratively increasing the number of temporally adjacent data points until the updated local clustering degree difference is less than or equal to the difference threshold, thereby obtaining the final number of temporally adjacent data points; and calculating the maximum number of extracted data points compared to the final temporally adjacent data points. Euclidean distance is used as the merging distance threshold; each data point is treated as a separate cluster and its neighboring clusters are obtained, wherein the local clustering degree of the neighboring clusters is greater than that of the cluster; if the minimum Euclidean distance between a cluster and any of its neighboring clusters is less than the maximum value of the merging distance threshold of the data points within the cluster, then the cluster and the neighboring clusters are merged; otherwise, the cluster is regarded as a stable cluster and the iterative merging is stopped; based on the dispersion of data points within the stable clusters, an anomaly index is determined; the time corresponding to the data points of stable clusters with an anomaly index greater than a set threshold is taken as the time of an anomaly in the state of the injection preparation tank.
[0007] The core advantage of this invention lies in its ability to sensitively identify potential stage boundary points by calculating local clustering differences. Furthermore, by iteratively increasing the number of temporally adjacent data points, it dynamically and adaptively calculates the merging distance threshold for data points, enabling potential stage boundary points to acquire larger thresholds. This allows normal stage boundary points to merge with slightly more distant high-density main clusters, thereby significantly reducing the false alarm rate. It solves the problem that traditional methods cannot distinguish between stage boundaries and real anomalies, and significantly improves the monitoring accuracy in multi-stage processes.
[0008] Preferably, the local clustering degree satisfies the expression: In the formula, For the first Local clustering of data points; For the first The neighboring data points of the data point in the th data point are at the th... Standard deviation of data values in each dimension; The index value and total number of all data points; For all data points, the neighborhood data points at the th The maximum value among the standard deviations of the data values in each dimension; For dimension index values and total count; It is a natural exponential function.
[0009] This invention uses the standard deviation of neighborhood data points across all dimensions to measure the degree of dispersion of neighborhood data points in multiple dimensions, providing an accurate and reliable mathematical basis for subsequently distinguishing densely distributed normal operating condition data from sparsely distributed outlier points or stage boundary points.
[0010] Preferably, the difference in local clustering satisfies the expression: In the formula, For the first The difference in local clustering of individual data points; For the first The mean of the local clustering degree of all temporally adjacent data points of a data point; For the first The first data point Local clustering of neighboring time-series data points; This represents the index value and total number of adjacent data points in the time series.
[0011] This invention reflects the stability of data points in a time series by calculating the difference in local clustering of adjacent data points. This indicator can effectively identify whether a data point is in a stable phase, at a phase boundary, or an outlier, providing a key input for subsequent dynamic adjustment of the merging threshold.
[0012] Preferably, the difference threshold is obtained by: extracting process data points of normal batches after manual verification, calculating the local clustering difference of all data points, constructing a curve of local clustering difference changing with time, and extracting the local clustering difference corresponding to the inflection point on the curve as the difference threshold.
[0013] This invention provides an objective threshold setting method by extracting the inflection point of the local clustering difference curve of normal batch data as a threshold. This threshold can effectively distinguish between stable and unstable data states, avoid the subjectivity of manually setting thresholds, and enhance the robustness of the method.
[0014] Preferably, the method for obtaining the neighboring clusters includes: calculating the average local clustering degree of all data points within any cluster as the local clustering degree of the cluster; and extracting clusters with local clustering degrees greater than the local clustering degree of the cluster as the neighboring clusters of the cluster.
[0015] Preferably, the anomaly index satisfies the expression: In the formula, For the first Anomaly index of a stable cluster; For the first The number of data points within a stable cluster; The total number of all data points; For the first The mean Euclidean distance between data points and cluster centers within a stable cluster; It represents the maximum value of the mean Euclidean distance between data points and cluster centers within all stable clusters.
[0016] This invention combines the size and internal dispersion of stable clusters. This index can effectively amplify the anomaly index of clusters with few and scattered data points, distinguishing them from normal clusters with large and concentrated data, thus achieving accurate identification of abnormal states.
[0017] Preferably, the method for obtaining the neighborhood data points includes: in a space composed of multi-dimensional data, taking any data point as the center, extracting all data points within a preset radius as the neighborhood data points of that data point.
[0018] Preferably, the method for obtaining temporally adjacent data points includes: for any data point, extracting data points that are temporally adjacent to the data point as temporally adjacent data points, and setting the initial number of temporally adjacent data points to 10.
[0019] Preferably, the collection of multi-dimensional data from the injection preparation tank at various times as data points includes: performing maximum and minimum normalization on the data values in each dimension at each time, and using the processed multi-dimensional data values as data points at each time; the multi-dimensional data refers to temperature, motor current, and pH value.
[0020] Secondly, the present invention provides an intelligent production monitoring system for injectable solutions, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned intelligent production monitoring method for injectable solutions is implemented.
[0021] By adopting the above technical solution, a computer program for intelligent production monitoring of the above-mentioned injection solution is generated and stored in a memory so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and the processor for convenient use.
[0022] The beneficial effects of this invention are as follows: (1) This invention dynamically adjusts the number of temporally adjacent data points by introducing local clustering differences, so that it can dynamically determine whether the data points are in a stable stage or in an unstable stage boundary or abnormal state. (2) Based on the number of neighboring data points in the final time series after the above dynamic adjustment, the present invention calculates the maximum Euclidean distance of data points within this range as the merging distance threshold of data points. Unstable points at the stage boundary will obtain a larger merging distance threshold, allowing them to merge with the high-density main clusters that are slightly farther away, ensuring that they will not be mistakenly isolated as anomalous points. This solves the core problem of traditional density clustering misreporting normal stage transitions as anomalous, and significantly improves the accuracy of intelligent production monitoring of injection solutions. Attached Figure Description
[0023] Figure 1 This is a flowchart illustrating an intelligent production monitoring method for injectable solutions according to the present invention; Figure 2 A schematic diagram of normalized multi-dimensional data; Figure 3 This is a schematic diagram of the detection results of the present invention; Figure 4 This is a schematic diagram of the detection results of the traditional density clustering algorithm. Detailed Implementation
[0024] 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, not all, of the embodiments of the present invention. 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.
[0025] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0026] This invention discloses an intelligent production monitoring method for injectable solutions, referring to... Figure 1 This includes steps S1-S5: S1. Collect multi-dimensional data from the drug preparation tank at various times as data points.
[0027] It should be noted that during the preparation of the drug solution, the state inside the tank is the result of the coupling of multiple process parameters. Traditional single-point monitoring cannot reflect the true reaction state of the drug solution. For example, the dissolution or neutralization reaction of the active pharmaceutical ingredient will cause changes in the pH value and viscosity of the liquid. The viscosity change will be directly reflected in the load current of the stirring motor, and the reaction itself may also be accompanied by temperature changes. Therefore, it is necessary to collect data from multiple dimensions such as temperature, current and pH to provide a data basis for subsequent analysis of the changes between parameters.
[0028] Specifically, multi-dimensional process parameters of the drug preparation tank are collected at different times. These multi-dimensional parameter data include: real-time temperature data collected by a temperature sensor installed on the tank body or jacket; real-time motor current data collected by a sensor installed in the control cabinet of the stirring motor at the top of the preparation tank; and real-time pH value data collected by an online pH meter installed inside the preparation tank. For example, the sampling frequency of each sensor can be set to 1Hz, which can be adjusted by the implementer according to the actual situation.
[0029] Considering that the unit dimension and quantity range may affect the analysis results when analyzing multidimensional data, the data values of each dimension at the real-time acquisition time need to be processed by maximum and minimum normalization before storage to eliminate the influence of unit dimension; at the same time, the processed multidimensional data values at each time are used as the data points at each time.
[0030] For example, Figure 2 This is a schematic diagram of the normalized multi-dimensional data. The horizontal axis represents time, and the vertical axis represents the normalized values of each dimension. The blue curve represents the normalized temperature data, the orange curve represents the normalized motor current data, and the green curve represents the normalized pH value data. The data curves clearly show changes during three normal process stages, and all dimensions of the data in the 450-500 range experience drastic and rapid fluctuations, indicating an anomaly in this range.
[0031] At this point, all data points have been obtained.
[0032] S2. Determine the local clustering degree based on the changes in the neighboring data points of any data point in each dimension.
[0033] It should be noted that in a standard batch of drug preparation, the combination of process parameters should be similar and continuously changing most of the time. These normal data points will cluster together in multidimensional space, forming a high-aggregation region. Conversely, if there is raw material clumping that causes a sudden increase in load, or if there is an error in adding acid or alkali that causes a sudden change in pH, the corresponding abnormal data points will deviate from this dense region and appear as a low-aggregation region. Therefore, analyzing the local aggregation degree of each data point is the mathematical basis for distinguishing normal operating conditions.
[0034] Specifically, in a space composed of multi-dimensional data, all data points within a preset radius are extracted from any data point as the center and used as the neighborhood data points of that data point.
[0035] The preset radius defines the distribution of data points within the neighborhood of a data point. Considering that the data is scaled to the [0,1] interval after normalization, all data points fall within a cube with a side length of 1 in three-dimensional space. If the radius is too small, it is easy to not find other neighbors in a small neighborhood, which may lead to the target data point being incorrectly identified as a low-cluster area. Therefore, the present invention sets the radius to 0.2.
[0036] The local clustering degree is determined based on the changes in the data values of the neighboring data points of any data point across each dimension; the local clustering degree satisfies the expression:
[0037] In the formula, For the first Local clustering of data points; For the first The neighboring data points of the data point in the th data point are at the th... Standard deviation of data values in each dimension; The index value and total number of all data points; For all data points, the neighborhood data points at the th The maximum value among the standard deviations of the data values in each dimension; For dimension index values and total count; It is a natural exponential function.
[0038] in, This value reflects the degree of dispersion of multi-dimensional data within the neighborhood of the current data point. A smaller value indicates that the neighboring data points are very similar across all dimensions, meaning the local distribution of the current data point is more clustered; conversely, a larger value indicates that the neighboring data points are very loosely distributed in at least one dimension, meaning the local distribution of the current data point is more sparse. It should be noted that in actual production processes, due to process fluctuations... It is always greater than zero.
[0039] At this point, the local clustering degree of all data points has been obtained.
[0040] S3. Calculate the difference in local clustering degree between any data point and its temporal neighboring data points as the local clustering degree difference; based on the relationship between the local clustering degree difference and the difference threshold, obtain the final number of temporal neighboring data points; calculate the maximum Euclidean distance between any data point and the final temporal neighboring data points as the merging distance threshold.
[0041] It should be noted that using the relative distance within the neighborhood radius can more accurately assess the distribution density of data points in multidimensional space. Traditional density clustering algorithms can effectively identify differences in density distribution in the data, thereby enabling outlier monitoring. However, actual drug preparation processes involve data changes at different stages. If clustering is not performed to distinguish the characteristics of changes at each stage, inaccurate clustering results can easily occur due to cross-stage issues. If the temporal neighborhood of a target data point contains anomalies or changes at different stages, causing instability, the local clustering degree of data points in the target data point's temporal neighborhood will differ significantly from the mean local clustering degree of data points in its temporal neighborhood. If the temporal neighborhood of a target data point represents a stable change at the same stage, there will be smaller differences, requiring an expansion of the neighborhood to capture the boundaries of stable stages. By constructing clustering indicators based on the distance of the adjusted temporal neighborhood of the target data point, the rationality of the clustering can be ensured.
[0042] Specifically, for any data point, an initial number of temporally adjacent data points is preset, and data points that are temporally adjacent to the data point are extracted as temporally adjacent data points of the data point. The selection of temporally adjacent data points takes into account the temporal continuity during the process, and it is more reliable to analyze whether the data changes in the time series are from the same stage. Therefore, by taking a smaller preset value for analysis, the stage changes can be refined. In this embodiment of the invention, the initial number of temporally adjacent data points is preset to 10.
[0043] The local clustering difference is determined based on the changes in the local clustering of all time-series neighboring data points for any given data point; the local clustering difference satisfies the expression:
[0044] In the formula, For the first The difference in local clustering of individual data points; For the first The mean of the local clustering degree of all temporally adjacent data points of a data point; For the first The first data point Local clustering of neighboring time-series data points; This represents the index value and total number of adjacent data points in the time series.
[0045] in, Reflecting the The degree of difference in local clustering of temporal neighboring data points for a given data point. The larger the value, the greater the difference in local clustering of temporal neighboring data points for that data point, which means that the temporal neighborhood of that data point is more unstable and more likely to be at the stage boundary or an outlier. Conversely, it means that the temporal neighborhood of the current data point is more stable and more likely to belong to the same stage.
[0046] Considering that in the normal drug preparation process, most data points in the stable stage have relatively high and concentrated local clustering due to the relatively stable parameter values of neighboring points, while a few data points at the stage boundary have significantly lower local clustering values for all time-series neighboring data points due to the unstable parameter values of neighboring points across stages; if the local clustering during the normal preparation process is plotted on a curve, there will be an inflection point, which is the dividing line between stability and instability.
[0047] Extract process data points from normal batches after manual verification, calculate the local clustering difference of all data points, construct a curve of local clustering difference changing over time, and extract the local clustering difference corresponding to the inflection point on the curve as the difference threshold.
[0048] It should be further explained that if a situation occurs during actual monitoring where the difference threshold is greater than the threshold, it indicates that the temporal neighbor data points of the current monitoring point are more unstable, which means that they may have crossed different stages or contain anomalies. In this case, increasing the number of temporal neighbor data points can ensure that a large merging threshold can be obtained, allowing merging with high-density points that are relatively far away, which helps to absorb boundary points into the main cluster to which they should belong; conversely, merging with high-density points that are very close to the point is only allowed, which helps to maintain the compactness of clusters in the same stage.
[0049] Specifically, if the first If the local clustering difference of a data point is greater than a local threshold, the number of time series data points is incremented by 1, and this process is repeated until the nth data point is reached. When the difference in local clustering degree of the updated data point, calculated after considering the number of its temporally neighboring data points, is less than or equal to a local threshold, the result is obtained. The final number of time-series neighbor data points for each data point; [This is followed by a list of all final time-series neighbor data points and their corresponding data points.] The maximum Euclidean distance between the nth data points is used as the nth... The threshold for merging distances between data points.
[0050] At this point, the merged distance threshold for all data points has been obtained.
[0051] S4. Treat each data point as a separate cluster and obtain its neighboring clusters. If the minimum Euclidean distance between a cluster and any of its neighboring clusters is less than the maximum value of the merging distance threshold for data points within that cluster, then merge that cluster with its neighboring clusters. Otherwise, treat the cluster as a stable cluster and stop iterative merging.
[0052] It should be noted that the introduction of the merging distance threshold makes the density clustering algorithm more flexible, enabling it to adjust to the dynamic changes of data points, thereby effectively capturing the changes and potential abnormal patterns in each stage of the drug preparation process and improving the accuracy of clustering.
[0053] Specifically, each data point is treated as a separate cluster, and the neighboring clusters of each cluster are obtained. The neighboring clusters are obtained by: calculating the average local clustering of all data points in any cluster as the local clustering of that cluster; traversing the local clustering of all clusters, and extracting the clusters with a local clustering greater than that of the cluster as the neighboring clusters.
[0054] The minimum Euclidean distance between any cluster and each of its neighboring clusters is calculated using the minimum distance method, which is a well-known technique and will not be elaborated upon here.
[0055] If the minimum Euclidean distance between any cluster and any of its neighboring clusters is less than the maximum value of the merging distance threshold for data points within that cluster, then that cluster is merged with that neighboring cluster; otherwise, that cluster is considered a stable cluster, and the iterative merging process is stopped.
[0056] For example, Figure 3 This is a schematic diagram of the detection results of the present invention, where the X-axis represents temperature data, the Y-axis represents motor current data, the Z-axis represents pH value data, and the normal clusters 0, 1, and 2 correspond to... Figure 2 In the three normal process stages, the red scatter dots represent the identified abnormal data points, corresponding to... Figure 2 The method of this invention successfully distinguishes normal process data into three different stages and accurately identifies abnormal data points, solving the problem that traditional methods have difficulty distinguishing stage boundaries. Figure 4 The diagram shows the detection results of a traditional density clustering algorithm. The diagram cannot distinguish between different process stages under normal operating conditions, and it is difficult to effectively distinguish between normal stage boundary data and real process anomaly data, resulting in a large number of false alarms and thus affecting the accuracy of monitoring.
[0057] At this point, all stable clusters and their corresponding cluster centers have been obtained.
[0058] S5. Determine the anomaly index based on the dispersion of data points within stable clusters; take the time corresponding to the data points of stable clusters with an anomaly index greater than the set threshold as the time when the injection preparation tank is in an abnormal state.
[0059] It should be noted that by adjusting the above clustering strategy, we can ensure that the process data points can be better distinguished between normal and abnormal changes. Considering that normal change data points are relatively concentrated in space and there are many data points, while abnormal changes often cause fluctuations in multiple parameters, are spatially dispersed and have a smaller proportion of data points; therefore, based on this characteristic, we calculate the degree of abnormality of each cluster, thereby extracting the abnormal moment and helping staff find the source of the problem.
[0060] Specifically, for any stable cluster, an anomaly index is determined based on the number and distribution of data points within that cluster; the anomaly index satisfies the expression:
[0061] In the formula, For the first Anomaly index of a stable cluster; For the first The number of data points within a stable cluster; The total number of all data points; For the first The mean Euclidean distance between data points and cluster centers within a stable cluster; It represents the maximum value of the mean Euclidean distance between data points and cluster centers within all stable clusters.
[0062] in, Reflecting the The size of a stable cluster; the larger this value, the fewer data points the stable cluster contains, which is consistent with the characteristics of anomalies. Reflecting the The anomaly index reflects the dispersion of data points within a stable cluster. A larger value indicates a more dispersed distribution of data points within the stable cluster, consistent with the characteristic of continuously discrete changes in anomalous parameters. If the stable cluster has few data points and high dispersion, its anomaly index will increase accordingly, making it more likely that the stable cluster is an anomalous cluster. It should be noted that in actual data distribution, data points will not completely overlap with the cluster centers. It is always greater than zero.
[0063] In response to stable clusters whose anomaly index exceeds a set threshold, data points within these stable clusters are identified as anomalous data points. The time corresponding to the anomalous data point is the time when the injection preparation tank is in an abnormal state. Corresponding control and early warning program instructions are generated to issue an early warning and notify relevant personnel to check process parameters.
[0064] The specific value of the threshold can be set according to the actual application scenario and needs. In this invention, the threshold is set to 0.6.
[0065] This invention also discloses an intelligent production monitoring system for injectable solutions, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an intelligent production monitoring method for injectable solutions according to this invention.
[0066] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.
Claims
1. A method for intelligent production monitoring of injectable solutions, characterized in that, include: Multi-dimensional data from the injection preparation tank at various times were collected as data points. For any data point, determine the local clustering degree based on the changes of its neighboring data points in each dimension; calculate the difference in local clustering degree of its temporal neighboring data points as the local clustering degree difference; in response to the local clustering degree difference being greater than the difference threshold, iteratively increase the number of temporal neighboring data points until the updated local clustering degree difference is less than or equal to the difference threshold, and obtain the final number of temporal neighboring data points; calculate the maximum Euclidean distance between the extracted data point and the final temporal neighboring data points as the merging distance threshold. Each data point is treated as a separate cluster, and its neighboring clusters are obtained. The local clustering degree of the neighboring clusters is greater than that of the cluster. If the minimum Euclidean distance between a cluster and any of its neighboring clusters is less than the maximum value of the merging distance threshold of the data points within the cluster, then the cluster and the neighboring clusters are merged. Otherwise, the cluster is regarded as a stable cluster, and the iterative merging is stopped. An anomaly index is determined based on the dispersion of data points within stable clusters. The time corresponding to the data points of stable clusters with an anomaly index greater than a set threshold is taken as the time when the injection preparation tank is in an abnormal state.
2. The intelligent production monitoring method for injectable solutions according to claim 1, characterized in that, The local clustering degree satisfies the expression: ； In the formula, For the first Local clustering of data points; For the first The neighboring data points of the data point in the th data point are at the th... Standard deviation of data values in each dimension; The index value and total number of all data points; For all data points, the neighborhood data points at the th The maximum value among the standard deviations of the data values in each dimension; For dimension index values and total count; It is a natural exponential function.
3. The intelligent production monitoring method for injection solutions according to claim 1, characterized in that, The difference in local clustering satisfies the expression: ； In the formula, For the first The difference in local clustering of individual data points; For the first The mean of the local clustering degree of all temporally adjacent data points of a data point; For the first The first data point Local clustering of neighboring time-series data points; This represents the index value and total number of adjacent data points in the time series.
4. The intelligent production monitoring method for injectable solutions according to claim 1, characterized in that, The difference threshold is obtained by extracting process data points of normal batches after manual verification, calculating the local clustering difference of all data points, constructing a curve of local clustering difference changing with time, and extracting the local clustering difference corresponding to the inflection point on the curve as the difference threshold.
5. The intelligent production monitoring method for injectable solutions according to claim 1, characterized in that, The method for obtaining the nearest neighbor clusters includes: Calculate the average local clustering degree of all data points within any cluster, and use it as the local clustering degree of that cluster; extract the clusters with local clustering degrees greater than the local clustering degree of that cluster, and use them as the neighboring clusters of that cluster.
6. The intelligent production monitoring method for injectable solutions according to claim 1, characterized in that, The anomaly index satisfies the expression: ； In the formula, For the first Anomaly index of a stable cluster; For the first The number of data points within a stable cluster; The total number of all data points; For the first The mean Euclidean distance between data points and cluster centers within a stable cluster; It represents the maximum value of the mean Euclidean distance between data points and cluster centers within all stable clusters.
7. The intelligent production monitoring method for injectable solutions according to claim 1, characterized in that, The method for obtaining the neighborhood data points includes: in a space composed of multi-dimensional data, taking any data point as the center, extracting all data points within a preset radius as the neighborhood data points of that data point.
8. The intelligent production monitoring method for injectable solutions according to claim 1, characterized in that, The method for obtaining temporally adjacent data points includes: for any data point, extracting data points that are temporally adjacent to the data point as temporally adjacent data points, and setting the initial number of temporally adjacent data points to 10.
9. The intelligent production monitoring method for injectable solutions according to claim 1, characterized in that, The process of collecting multi-dimensional data from the injection preparation tank at various times as data points includes: performing maximum and minimum normalization on the data values of each dimension at each time, and using the processed multi-dimensional data values as data points at each time; the multi-dimensional data refers to temperature, motor current, and pH value.
10. An intelligent production monitoring system for injection solutions, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement an intelligent production monitoring method for an injectable solution according to any one of claims 1-9.