A method for predicting the production quality of an alginate sodium hyaluronate anti-HPV gel

Abstract

The application discloses a production quality prediction method of alginate sodium hyaluronate anti-HPV gel and belongs to the technical field of quality prediction management, and specifically comprises the following steps: inputting an instruction single number; obtaining a stirring shaft forward and reverse rotation operation record in a cross-linking stirring stage from production process records to form a forward rotation and reverse rotation load sequence; extracting a load change rate and a recovery time length before and after each switching, combining the load change rate and the recovery time length into a switching process response feature and arranging the switching process response feature into a batch process feature sequence; calculating a process feature deviation degree of a target batch and historical batches and screening a reference batch set; determining a fusion weight according to the process feature deviation degree and a production time interval, and obtaining a quality prediction value through weighted calculation; and storing the prediction record in a database and sending the prediction record to a release audit node. The application can provide batch-specific quality prediction data for batch release audit in the production stage.

Description

Technical Field

[0001] This invention relates to the field of quality prediction and management technology, specifically to a method for predicting the production quality of alginate sodium hyaluronate anti-HPV gel. Background Technology

[0002] A gel dressing formed by the cross-linking reaction of alginate and sodium hyaluronate has been applied in the field of anti-human papillomavirus (HPV) biological barrier therapy. The production process of this type of gel product typically includes raw material dissolution, cross-linking stirring, filling, and packaging. The cross-linking stirring stage is the key step in determining the microstructure of the gel's network. During production, different batches of alginate raw materials of the same product specification exhibit inherent differences in the source of extracted seaweed and the range of molecular weight distribution. Similarly, different batches of sodium hyaluronate raw materials show batch-to-batch fluctuations in molecular weight and degree of substitution. These differences in the combination of raw materials from different batches manifest as different response patterns of the materials to stirring during the cross-linking stirring process. After batch completion, the manufacturer needs to assess the anti-HPV activity of the finished gel to provide a basis for batch release approval.

[0003] In existing production management processes, the testing of the anti-HPV activity of the finished gel is usually completed by external laboratories through in vitro cell experiments. The complete testing cycle, from sample delivery, cell culture, virus inoculation to titer determination, spans multiple working days, and the time for obtaining test results lags far behind the production completion time. Some manufacturers attempt to make empirical inferences about batch quality based on the viscosity grade of raw materials, crosslinking agent dosage parameters, or the average power value during the crosslinking stirring stage. Other manufacturers simply average the test values ​​of historical batches of the same specification as a reference estimate for the new batch. However, the static indicators of raw materials cannot reflect the unique dynamic response differences of different batches of raw material combinations during the crosslinking process. The average power value or steady-state torque value during the crosslinking stirring stage averages the transition process information at the moment of switching the stirring direction, compressing the key dynamic characteristics reflecting the differences in raw material combinations. Furthermore, simply averaging the test values ​​of historical batches ignores the time correlation between different historical batches and the target batch in terms of asymptotic drift of raw material properties, resulting in manufacturers lacking batch-specific quality prediction data to support their batch release review. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the production quality of alginate-sodium hyaluronate anti-HPV gel, thereby solving the following technical problems:

[0005] Existing methods use static raw material indicators or average crosslinking stirring power to infer quality, averaging the information of the transition process at the moment of switching stirring direction, compressing the dynamic characteristics reflecting the differences in raw material combinations, and ignoring the time correlation between historical batches and target batches in terms of asymptotic drift of raw material properties, thus failing to provide batch-specific quality prediction data support.

[0006] The objective of this invention can be achieved through the following technical solutions: A method for predicting the production quality of alginate-sodium hyaluronate anti-HPV gel includes the following steps: S1. Input the instruction order number of the target production batch, which is associated with the batch number of alginate raw material, sodium hyaluronate raw material, crosslinking agent, reactor identification, production instruction issuance time, and production process record; S2. Obtain the operation records of the stirring shaft rotating forward and backward sequentially during the crosslinking stirring stage from the production process records, and form forward rotation load sequence and reverse rotation load sequence respectively; S3. Extract the load change rate and recovery time before and after each forward / reverse switch, combine them into switching process response features, and arrange them into a batch process feature sequence according to the switching order; S4. Retrieve the process feature sequences of historical batches from the same manufacturer from the historical batch database, calculate the deviation of the process features between the target batch and each historical batch, and select a set of reference batches based on the deviation of the process features. S5. Obtain the anti-HPV activity test value and production date of the reference batch of finished products. Determine the fusion weight by the process characteristic deviation and the interval between the production date of the reference batch and the production instruction of the target batch. Calculate the quality prediction value by weighting. S6. Combine the quality prediction value with the instruction sheet number and the production instruction issuance time to form a prediction record and store it in the historical batch database. Send the reference batch instruction sheet number and quality prediction value to the release review node.

[0007] As a further aspect of the present invention: the specific processes for forming the forward load sequence and the reverse load sequence in S2 are as follows: The product specification code in the instruction sheet number is analyzed, and the forward rotation duration, reverse rotation duration, and direction pause duration are retrieved from the mixing cycle configuration table based on the product specification code. The cross-linking mixing stage consists of alternating forward rotation, pause, and reverse rotation duration segments. During the pause segment, the mixing shaft drive input is zero. Within each forward rotation duration segment, the load value of the drive mechanism is recorded at a fixed acquisition interval. The load values ​​of all forward rotation duration segments are arranged sequentially by acquisition time to form a forward rotation load sequence. Within each reverse rotation duration segment, the load value of the drive mechanism is recorded at the same fixed acquisition interval. The load values ​​of all reverse rotation duration segments are arranged sequentially by acquisition time to form a reverse rotation load sequence.

[0008] As a further aspect of the present invention: the specific process of extracting the load change rate and recovery time before and after each forward / reverse switching in S3 is as follows: The switching time from forward to reverse rotation is taken as the first switching point. A first number of load values ​​are extracted along the reverse direction of the forward load sequence as the first leading sequence. A second number of load values ​​are extracted along the forward direction of the reverse load sequence from the first switching point as the first following sequence. The average difference between adjacent load values ​​is calculated using the load values ​​at the end of the first leading sequence. The average difference is divided by the sampling interval, and the result is taken as the load change rate from forward to reverse rotation. The average load value of the stable segment far from the switching area in the reverse load sequence is superimposed with a preset fluctuation tolerance to form a steady-state recovery interval. The sampling time when the first load value falls into the steady-state recovery interval is located in the first following sequence. The time span between the sampling time and the first switching point is calculated as the recovery time from forward to reverse rotation. The switching time from reverse to forward rotation is taken as the second switching point, and the load change rate and recovery time from reverse to forward rotation are calculated in the same way.

[0009] As a further aspect of the present invention: In step S3, the specific process of combining the load change rate and recovery time into a switching process response feature and arranging them into a batch process feature sequence according to the switching order is as follows: Obtain the baseline load change rate and baseline recovery time associated with the product specification code in the instruction sheet number. Divide the load change rate of each forward-to-reverse switch by the baseline load change rate, and the resulting ratio is taken as the first change rate ratio. Divide the recovery time of the forward-to-reverse switch by the baseline recovery time, and the resulting ratio is taken as the first recovery time ratio. Combine the first change rate ratio and the first recovery time ratio to form the switching process response characteristics of the forward-to-reverse switch. Convert the load change rate and recovery time of each reverse-to-forward switch into the second change rate ratio and the second recovery time ratio in the same way, and combine them to form the switching process response characteristics of the reverse-to-forward switch. Arrange all switching process response characteristics of the crosslinking stirring stage in the order of the switching occurrence to form a batch process characteristic sequence.

[0010] As a further aspect of the present invention: in step S4, the specific process for calculating the deviation of the process characteristics between the target batch and each historical batch is as follows: The batch process feature sequence of the target batch is recorded as the target sequence, and the batch process feature sequence of the historical batch is recorded as the comparison sequence. The target sequence is subjected to segmentation point detection. The switching point where the absolute value of the difference between the first change rate ratios of the response features of adjacent switching processes exceeds the preset change rate segmentation limit is marked as the segmentation point. The target sequence is divided into several target sequence segments using the segmentation points. According to the position of the segmentation point in the target sequence, the comparison sequence is divided into several comparison sequence segments at the same position in the comparison sequence. Align each target sequence segment with the corresponding comparison sequence segment position by position, calculate the sum of the squared difference between the first rate of change ratio and the squared difference between the first recovery time ratio at the aligned positions, and use the sum as the single-point difference value of the position. Accumulate all single-point difference values ​​between all target sequence segments and the corresponding comparison sequence segments, and use the accumulated result as the process characteristic deviation degree between the target batch and the historical batch.

[0011] As a further aspect of the present invention: the specific process of selecting the reference batch set based on the process characteristic deviation in S4 is as follows: All historical batches participating in the comparison are sorted in ascending order of process feature deviation value to form a deviation sequence. The difference in process feature deviation between adjacent positions in the deviation sequence is calculated, and the larger process feature deviation in the group of adjacent positions with the largest difference is taken as the critical deviation. All historical batches in the deviation sequence whose process feature deviation is less than the critical deviation are extracted as a reference batch set.

[0012] As a further aspect of the present invention: In step S5, the specific process for determining the fusion weight based on the process characteristic deviation and the interval between the reference batch production date and the target batch production instruction issuance time is as follows: For each reference batch in the reference batch set, a similarity score is calculated using the negative value of the process characteristic deviation of the reference batch as the exponent, with the natural constant as the base. The number of days between the production date of the reference batch and the production instruction time of the target batch is divided by a preset half-life to obtain the decay index. A time decay factor is calculated using the decay index as the exponent, with 0.5 as the base. The similarity score of the reference batch is multiplied by the time decay factor to obtain the fusion coefficient. The sum of the fusion coefficients of all reference batches in the reference batch set is calculated. The fusion coefficient of each reference batch is divided by the sum of the fusion coefficients to obtain the fusion weight of the reference batch. The process of weighted calculation to obtain the quality prediction value is as follows: the fusion weight of each reference batch is multiplied by the finished anti-HPV activity test value of the reference batch to obtain the single-batch weighted value. The single-batch weighted values ​​of all reference batches in the reference batch set are added together, and the sum is used as the quality prediction value.

[0013] As a further aspect of the present invention: In step S6, the specific process of combining the quality prediction value with the instruction order number and the production instruction issuance time to form a prediction record and storing it in the historical batch database is as follows: Extract the product code field, production date field, and batch sequence field from the instruction order number, and combine them with the quality prediction value and the production instruction issuance time to form a fixed-length record entry. Retrieve the release control threshold range corresponding to the product code field, and compare the quality prediction value with the release control threshold range. If the quality prediction value falls within the release control threshold range, a pending release label is generated; if the quality prediction value falls outside the release control threshold range, a pending review label is generated. Combine the fixed-length record entry, label, and prediction generation time to form a prediction record, and store the prediction record in the historical batch database. The instruction sheet number and process characteristic deviation of each reference batch in the reference batch set are arranged into list items. All list items are combined into a list, and the list and prediction records are sent to the release review node.

[0014] The beneficial effects of this invention are: This invention extracts the load change rate and recovery time before and after each forward and reverse rotation from the operation record of the stirring shaft during the crosslinking stirring stage. These are combined into switching process response features and arranged into a batch process feature sequence according to the switching order. This completely preserves the transition process information of the material to the switching of stirring direction, allowing the inherent dynamic differences between different raw material batches in the crosslinking process to be preserved in sequence form. By detecting sequence segmentation points and comparing each position sequentially, the deviation of the process features between the target batch and each historical batch is calculated. A reference batch set is selected using the deviation threshold. The similarity score is then calculated using the negative value of the process feature deviation as the exponent of the natural constant. Simultaneously, a time decay factor is constructed by introducing the ratio of the number of days between the reference batch's production date and the target batch's production instruction issuance time to a preset half-life. The similarity score is multiplied by the time decay factor and normalized to obtain a fusion weight. This weighted fusion of the reference batch's finished product anti-HPV activity test value yields a quality prediction value. Thus, batch-specific quality prediction data can be provided for batch release review nodes during the production stage. Attached Figure Description

[0015] The invention will now be further described with reference to the accompanying drawings.

[0016] Figure 1 This is a flowchart illustrating the present invention. Detailed Implementation

[0017] 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.

[0018] Please see Figure 1 As shown, this invention provides a method for predicting the production quality of alginate-sodium hyaluronate anti-HPV gel, comprising the following steps: S1. Input the instruction order number of the target production batch, which is associated with the batch number of alginate raw material, sodium hyaluronate raw material, crosslinking agent, reactor identification, production instruction issuance time, and production process record; S2. Obtain the operation records of the stirring shaft rotating forward and backward sequentially during the crosslinking stirring stage from the production process records, and form forward rotation load sequence and reverse rotation load sequence respectively; S3. Extract the load change rate and recovery time before and after each forward / reverse switch, combine them into switching process response features, and arrange them into a batch process feature sequence according to the switching order; S4. Retrieve the process feature sequences of historical batches from the same manufacturer from the historical batch database, calculate the deviation of the process features between the target batch and each historical batch, and select a set of reference batches based on the deviation of the process features. S5. Obtain the anti-HPV activity test value and production date of the reference batch of finished products. Determine the fusion weight by the process characteristic deviation and the interval between the production date of the reference batch and the production instruction of the target batch. Calculate the quality prediction value by weighting. S6. Combine the quality prediction value with the instruction sheet number and the production instruction issuance time to form a prediction record and store it in the historical batch database. Send the reference batch instruction sheet number and quality prediction value to the release review node.

[0019] In a preferred embodiment of the present invention, the specific process of forming the forward load sequence and the reverse load sequence in step S2 is as follows: After a production order is issued, the production management equipment reads the order number of the target production batch. The order number is a string of letters and numbers, embedding the product specification code, production date, and batch serial number. For example, in the order number "HPV20240513003", the first three digits "HPV" represent the product specification code, the middle eight digits "20240513" represent the production date, and the last three digits "003" represent the batch serial number for that day. After extracting the product specification code "HPV", the production management equipment accesses a pre-built mixing cycle configuration table. The mixing cycle configuration table is a data table stored in the production management equipment's memory. Each record in the table corresponds to a product specification code, and the record content includes the required forward rotation duration, reverse rotation duration, and turning pause duration for that product specification. For the product specification code "HPV", the mixing cycle configuration table records a forward rotation duration of 5 seconds, a reverse rotation duration of 5 seconds, and a turning pause duration of 0.6 seconds.

[0020] After the crosslinking and stirring stage begins, the stirring shaft performs alternating rotational motion. First, it enters a forward rotation phase, where the stirring shaft rotates in the first direction for 5 seconds, and the drive mechanism transmits torque to the stirring shaft via a coupling. During this forward rotation phase, the torque sensor outputs the real-time torque value of the drive mechanism at a fixed sampling interval of 0.04 seconds. Each torque value represents the load on the stirring bearing at that sampling moment. The 125 torque values ​​collected within 5 seconds are arranged in ascending order of sampling time, forming a load sequence within the forward rotation phase. After the forward rotation phase ends, the stirring shaft enters a pause phase for 0.6 seconds. During this period, the drive mechanism stops outputting torque, the torque sensor reading returns to zero, and no load value is recorded during the pause phase. After the pause phase ends, it enters a reverse rotation phase, where the stirring shaft rotates in the second direction opposite to the first direction for 5 seconds. The torque sensor again collects torque values ​​at 0.04-second intervals, and the 125 torque values ​​collected within 5 seconds are arranged in sampling order, forming a load sequence within the reverse rotation phase. After the reverse rotation period ends, the pause period begins again, and this cycle repeats until the crosslinking stirring stage ends. The entire crosslinking stirring stage consists of 30 forward rotation periods and 30 reverse rotation periods.

[0021] The 30 load sequences within the forward rotation segments are concatenated sequentially, with the last torque value of the first forward rotation segment immediately followed by the first torque value of the second forward rotation segment, and so on, forming a forward rotation load sequence containing 3750 torque values. Similarly, the 30 load sequences within the reverse rotation segments are concatenated sequentially to form a reverse rotation load sequence also containing 3750 torque values. This concatenation operation does not perform any interpolation or filtering on the torque values; it only performs sequential splicing to completely preserve the original transition details of the load before and after each switch. In addition to torque values, the load values ​​for the drive mechanism can also be motor current or power values ​​that are proportional to the torque. This embodiment uses torque values ​​as an example.

[0022] In another preferred embodiment of the present invention, the specific process of extracting the load change rate and recovery time before and after each forward / reverse switching in step S3 is as follows: After the forward and reverse load sequences are formed, the production management equipment begins to extract the load change characteristics corresponding to each change in stirring direction. First, the switching point is located. The end time of each forward rotation duration can be extracted from the production process record; this time is the switching point from forward to reverse rotation. Taking the first switching point from forward to reverse rotation as an example, the timestamp corresponding to this switching point is the end of the 5th second after the start of the crosslinking stirring stage. This moment is recorded as the first switching point.

[0023] Using the first switching point as a reference, 20 torque values ​​are extracted in the reverse direction of the forward load sequence. These 20 torque values ​​are recorded in the 20 most recent sampling moments before the switching point and are arranged chronologically to form the first leading sequence. These 20 torque values ​​reflect the gradual load decay process at the end of the forward rotation. Then, 60 torque values ​​are extracted in the forward direction of the reverse load sequence, starting from the first switching point. These 60 torque values ​​are recorded in the first 60 sampling moments after the switching point and are arranged chronologically to form the first following sequence. These 60 torque values ​​reflect the load recovery process from zero at the initial stage of reverse startup.

[0024] When calculating the load change rate from forward to reverse rotation, the eight most recent torque values ​​from the end of the first leading sequence are taken, as these eight torque values ​​are closest to the switching point. The difference between any two adjacent torque values ​​is calculated sequentially (subtracting the previous torque value from the next), resulting in seven differences. The absolute values ​​of these differences are averaged and then divided by the sampling interval of 0.04 seconds. The quotient is the load change rate from forward to reverse rotation. For example, if the average of the seven differences is -0.12 N·m, then the load change rate is 0.12 divided by 0.04, which equals 3.0 N·m per second. This value represents the rate of torque decay before the switching.

[0025] When calculating the recovery time from forward to reverse rotation, it is necessary to first determine the average load value of the stable segment of the reverse load sequence. A portion of the reverse load sequence far from the switching area is selected, specifically from the 200th torque value to the end of the sequence. This segment is sufficiently far from the switching influence zone. The arithmetic mean of the torque values ​​within this segment is calculated to obtain the average load value of the stable segment. A preset fluctuation tolerance is set to ±5% of the average load value of the stable segment. The average load value of the stable segment is multiplied by 1.05 as the upper limit of the steady-state recovery interval, and multiplied by 0.95 as the lower limit, forming the steady-state recovery interval. Then, in the first following sequence, starting from the first torque value, iterates backward, checking whether each torque value falls within the steady-state recovery interval. When a torque value falls into this interval for the first time, the sampling time corresponding to that torque value is recorded. The time span between this sampling time and the first switching point is calculated; this time span is the recovery time from forward to reverse rotation. For example, the average load during the stable phase is 4.8 N·m, and the steady-state recovery range is from 4.56 N·m to 5.04 N·m. The 23rd torque value of 4.62 N·m in the first following sequence falls into the range for the first time. This sampling time is 0.92 seconds after the first switching point, so the recovery time is 0.92 seconds.

[0026] For the switching point from reverse to forward rotation, the end of the reverse duration is taken as the second switching point. The second leading sequence and the second following sequence are extracted using the same method as described above, with the same number of extractions: 20 and 60 respectively. The average difference between the last 8 torque values ​​in the second leading sequence is calculated and divided by the sampling interval to obtain the load change rate from reverse to forward rotation. Then, the average load of a stable segment far from the switching region in the forward rotation load sequence is calculated and superimposed with the same fluctuation tolerance to form a steady-state recovery interval. The first torque value falling into this interval in the second following sequence is located, and the time span between its sampling time and the second switching point is calculated as the recovery time from reverse to forward rotation. This process is repeated for all 60 switching operations in the crosslinking stirring stage, resulting in 30 sets of load change rates and recovery times from forward to reverse rotation, and 30 sets of load change rates and recovery times from reverse to forward rotation.

[0027] In another preferred embodiment of the present invention, the specific process of combining the load change rate and recovery time into a switching process response feature and arranging them into a batch process feature sequence according to the switching order in step S3 is as follows: For a given product specification code "HPV", the process parameter table stores the corresponding baseline load change rate and baseline recovery time. The baseline load change rate is an ideal reference value obtained from statistical analysis of a large amount of normal batch data under standard process conditions, for example, 2.5 N·m / s; the baseline recovery time is an ideal reference value for recovery time under standard process conditions, for example, 0.8 seconds. These two baseline values ​​can be retrieved from the storage device after the product specification code is determined.

[0028] For each forward-to-reverse switch, the previously calculated load change rate is divided by the baseline load change rate of 2.5, and the quotient is recorded as the first change rate ratio. For example, if the load change rate for a switch is 3.0 N·m / s, then the first change rate ratio is 1.2. The recovery time for this switch is divided by the baseline recovery time of 0.8, and the quotient is recorded as the first recovery time ratio. For example, if the recovery time for this switch is 0.92 seconds, then the first recovery time ratio is 1.15. The first change rate ratio of 1.2 and the first recovery time ratio of 1.15 are combined as the switch response characteristics for this forward-to-reverse switch. The combination method is to arrange the two ratios in chronological order, forming an ordered pair of values. The first position in this pair of values ​​is fixed to store the change rate ratio, and the second position is fixed to store the recovery time ratio.

[0029] For each reverse-to-forward switch, the second rate of change is obtained by dividing the load change rate by the baseline load change rate of 2.5; the second recovery time ratio is obtained by dividing the recovery time by the baseline recovery time of 0.8. The second rate of change ratio and the second recovery time ratio are combined in the same order to form the switching response characteristics of that reverse-to-forward switch. The baseline load change rate uses the same value for both forward-to-reverse and reverse-to-forward switch types, and the baseline recovery time also uses the same value because the movement of the stirring shaft in both directions is symmetrically designed.

[0030] The crosslinking stirring stage involves 60 switching operations, sequentially: the first switch from forward to reverse, the first switch from reverse to forward, the second switch from forward to reverse, the second switch from reverse to forward, and so on, up to the 30th switch from reverse to forward. Following the actual order of these switching operations, the response characteristics of each switching process are arranged sequentially, forming an ordered sequence of 60 elements. This sequence constitutes the batch process characteristic sequence for the target batch. Each element in the sequence is a combination of a rate of change ratio and a recovery time ratio. For example, the first element of the sequence represents the combination for the first switch from forward to reverse, with values ​​of 1.2 and 1.15; the second element represents the combination for the first switch from reverse to forward, with values ​​of 1.18 and 1.22; and so on. The batch process characteristic sequence fully records the dynamic response pattern of the material to repeated switching of stirring direction throughout the entire crosslinking stirring stage. This response pattern is determined by the specific combination of alginate, sodium hyaluronate, and crosslinking agent used in this batch. Different combinations of raw materials will produce different sequences for the same product specifications. The next step will be to compare this sequence with historical batches to assess the similarity between batches.

[0031] In another preferred embodiment of the present invention, the specific process of calculating the deviation of the process characteristics between the target batch and each historical batch in step S4 is as follows: The batch process feature sequence of the target batch has been generated in step S3. This sequence contains 60 switching process response features, each consisting of a first change rate ratio and a first recovery time ratio or a second change rate ratio and a second recovery time ratio, arranged in chronological order of switching occurrence. This batch process feature sequence of the target batch is designated as the target sequence. All historical production batches from the same manufacturer and under the same product specification code that have completed anti-HPV activity testing are retrieved from the historical batch database. Each historical production batch has had its own batch process feature sequence generated and stored in the same manner during the production process. The batch process feature sequence of a specific historical production batch currently being compared is designated as the comparison sequence.

[0032] First, segmentation point detection is performed on the target sequence. The purpose of segmentation point detection is to identify switching locations in the target sequence where the dynamic response pattern changes significantly, so that subsequent segment-by-segment comparisons can be performed within time intervals with similar response patterns. The response features of two adjacent switching processes in the target sequence are traversed, and the absolute value of the difference in the first rate of change ratio between adjacent features is checked sequentially. For example, if the 3rd and 4th switches in the target sequence are adjacent, the first rate of change ratio for the 3rd switch is 1.2, and the first rate of change ratio for the 4th switch is 1.52, with an absolute difference of 0.32. The preset rate of change segmentation threshold is set to 0.10, which is determined based on the maximum allowable difference in the rate of change ratio between adjacent features within the normal production fluctuation range for products of the same specification. Since 0.32 exceeds 0.10, the position of the 4th switch is marked as a segmentation point. This process continues to traverse all 59 pairs of adjacent switching features, and whenever the absolute value of the difference in the first rate of change ratio between adjacent features exceeds 0.10, a segmentation point is marked at the next switching location. After the traversal is completed, the target sequence is marked with split points at several switching positions. For example, there are 4 split points at the 4th, 15th, 38th and 52nd switching positions.

[0033] The target sequence is divided into five segments using these four segmentation points. Specifically, the first target sequence segment contains four handover response features from the first to the fourth handover; the second target sequence segment contains 11 features from the fifth to the fifteenth handover; the third target sequence segment contains 23 features from the sixteenth to the thirty-eighth handover; the fourth target sequence segment contains 14 features from the thirty-ninth to the fifty-second handover; and the fifth target sequence segment contains eight features from the fifty-third to the sixtyth handover. Each segmentation point belongs to the preceding sequence segment, and the first feature after a segmentation point belongs to the next sequence segment. The positions of these four segmentation points in the target sequence are recorded as 4th, 15th, 38th, and 52nd. Then, segmentation is performed at the corresponding positions in the comparison sequence according to these same positions, i.e., the comparison sequence is divided into five segments at the 4th, 15th, 38th, and 52nd handover positions in the comparison sequence. The number of switching features contained in each comparison sequence segment is exactly the same as that in the corresponding target sequence segment.

[0034] After segmentation, a segment-by-segment, position-by-position alignment and comparison are performed. Taking the first target sequence segment and the first comparison sequence segment as examples, each segment contains four handover process response features. The first positions of both segments are aligned, and the first change rate ratio (1.2) and first recovery duration ratio (1.15) of the first position in the target sequence segment are extracted. The first change rate ratio (1.08) and first recovery duration ratio (1.05) of the first position in the comparison sequence segment are extracted. The square of the difference in the first change rate ratio is calculated: 1.2 minus 1.08 equals 0.12, and the square of 0.12 is 0.0144. The square of the difference in the first recovery duration ratio is calculated: 1.15 minus 1.05 equals 0.10, and the square of 0.10 is 0.01. The sum of these two values ​​is 0.0244, which is the single-point difference value for the first position. The same calculation is performed for the second, third, and fourth positions, respectively, to obtain the single-point difference values ​​for each corresponding position. Then, the single-point difference values ​​of the four positions within the first target sequence segment are summed to obtain the intra-segment difference value of that sequence segment. The same position-by-position alignment and single-point difference value calculation are performed sequentially on the second to fifth target sequence segments and their corresponding comparison sequence segments. All single-point difference values ​​of all five sequence segments are summed, and the summed result is the process feature deviation between the target batch and the currently compared historical batch. This process feature deviation is then iterated through each historical batch in the historical batch database to obtain the process feature deviation between the target batch and each historical batch.

[0035] In another preferred embodiment of the present invention, the specific process of selecting the reference batch set based on the process characteristic deviation in step S4 is as follows: Assume there are 42 historical production batches under the same product specification code in the historical batch database, and 42 process characteristic deviation values ​​were calculated in the previous step. These values ​​are then sorted in ascending order to form a deviation sequence. For example, the first five values ​​in the sorted sequence are 0.34, 0.41, 0.52, 0.59, and 0.63, with a maximum value of 9.87.

[0036] Calculate the difference in process feature deviation between each pair of adjacent positions in the deviation sequence. That is, subtract the value of the first position from the value of the second position, subtract the value of the second position from the value of the third position, and so on, resulting in 41 differences. For example, the difference between 0.34 for the first position and 0.41 for the second position is 0.07, and the difference between 0.41 for the second position and 0.52 for the third position is 0.11. Assuming that after traversing all 41 differences, the largest difference is found between the 6th and 7th positions, with a process feature deviation of 0.98 for the 6th position and 2.35 for the 7th position, the difference being 1.37. The position of this largest difference represents the critical region where a historical batch highly similar to the target batch transitions to a historical batch with significantly reduced similarity. The larger process feature deviation value of 2.35 in this pair of adjacent positions is taken as the critical deviation. The critical deviation is a dynamically determined limit value that is adaptively generated based on the actual distribution of each batch of target sequences compared with all historical batches. Different target batches correspond to different critical deviations.

[0037] After determining the critical deviation, all 42 historical batches in the deviation sequence were iterated, and all historical batches with a process characteristic deviation less than 2.35 were extracted. Based on the example data above, the process characteristic deviations of the 1st to 6th historical batches in the deviation sequence were 0.34, 0.41, 0.52, 0.59, 0.63, and 0.98, respectively, all less than 2.35. Therefore, these six historical batches were extracted and formed into a reference batch set. The 7th and subsequent historical batches had process characteristic deviations greater than or equal to 2.35, showing a significant difference in crosslinking dynamic response pattern from the target batch, and were not included in the reference batch set. The historical batches in the reference batch set represent production cases with highly similar dynamic response patterns to the target batch during the crosslinking stirring process. Their finished product anti-HPV activity test values ​​will be used as a reference for weighted fusion in subsequent steps.

[0038] In another preferred embodiment of the present invention, the specific process of determining the fusion weight based on the process characteristic deviation and the interval between the reference batch production date and the target batch production instruction issuance time in step S5 is as follows: The reference batch set contains six historical production batches. Each historical batch has two key parameters: one is the process characteristic deviation calculated in step S4, and the other is the production date of that historical batch. The production date can be extracted from the instruction order number stored in the historical batch database. The production instruction issuance time for the target batch has been obtained from the instruction order number, for example, May 13, 2024. The process characteristic deviations of the six reference batches are 0.34, 0.41, 0.52, 0.59, 0.63, and 0.98, respectively, corresponding to production dates of May 10, 2024, May 8, 2024, April 25, 2024, April 12, 2024, March 28, 2024, and January 15, 2024, respectively.

[0039] First, the similarity score for each reference batch is calculated. Using the natural constant 2.71828 as the base and the negative value of the process characteristic deviation as the exponent, a power operation is performed. For the first reference batch, the process characteristic deviation is 0.34, which is taken as -0.34. 2.71828 raised to the power of -0.34 yields 0.7118, which is the similarity score for this reference batch. The similarity score measures the degree of similarity between the reference batch and the target batch in their crosslinking stirring dynamic response mode. The smaller the process characteristic deviation, the closer the similarity score is to 1.0, indicating that the dynamic responses of the two batches are more similar; the larger the deviation, the closer the similarity score is to 0, indicating a greater difference. The second reference batch has a deviation of 0.41, and the similarity score is 2.71828 raised to the power of -0.41, resulting in 0.6636. The third batch has a deviation of 0.52, and a similarity score of 0.5945. The fourth deviation is 0.59, and the similarity score is 0.5543. The fifth deviation is 0.63, and the similarity score is 0.5326. The sixth deviation is 0.98, and the similarity score is 0.3753.

[0040] Next, the time decay factor for each reference batch is calculated. The preset half-life is a pre-defined time parameter representing the number of days it takes for the reference batch's reference value for the target batch's quality prediction to decay to half. In this embodiment, the preset half-life is 30 days. For the first reference batch, its production date is May 10, 2024, and the target batch's production order was issued on May 13, 2024, a difference of 3 days. Dividing the day interval 3 by the preset half-life of 30 yields a decay index of 0.1. Using 0.5 as the base and 0.1 as the exponent, 0.5 raised to the power of 0.1 is calculated, resulting in 0.9330, which is the time decay factor for the first reference batch. The time decay factor reflects the impact of the gradual drift of raw material properties over time on the historical batch reference value. The closer the production date of a reference batch is to the target batch, the closer its time decay factor is to 1.0, and the more its contribution to quality prediction is retained. The second reference batch was produced on May 8th, 5 days after May 13th. Its decay index is 5 divided by 30, which equals 0.1667. The time decay factor is 0.5 raised to the power of 0.1667, resulting in 0.8909. The third reference batch was produced 18 days later, with a decay index of 0.6 and a time decay factor of 0.6598. The fourth reference batch was produced 31 days later, with a decay index of 1.0333 and a time decay factor of 0.4884. The fifth reference batch was produced 46 days later, with a decay index of 1.5333 and a time decay factor of 0.3457. The sixth reference batch was produced 119 days later, with a decay index of 3.9667 and a time decay factor of 0.0640.

[0041] The similarity score of each reference batch is multiplied by the time decay factor to obtain the fusion coefficient. For the first reference batch, 0.7118 multiplied by 0.9330 equals 0.6641. For the second, 0.6636 multiplied by 0.8909 equals 0.5913. For the third, 0.5945 multiplied by 0.6598 equals 0.3923. For the fourth, 0.5543 multiplied by 0.4884 equals 0.2707. For the fifth, 0.5326 multiplied by 0.3457 equals 0.1841. For the sixth, 0.3753 multiplied by 0.0640 equals 0.0240. The sum of the fusion coefficients for the six reference batches is 0.6641 + 0.5913 + 0.3923 + 0.2707 + 0.1841 + 0.0240 equals 2.1265. Dividing the fusion coefficient of each reference batch by the sum of the fusion coefficients yields the fusion weight. The first reference batch fusion weight is 0.6641 divided by 2.1265, which equals 0.3123; the second is 0.5913 divided by 2.1265, which equals 0.2781; the third is 0.3923 divided by 2.1265, which equals 0.1845; the fourth is 0.2707 divided by 2.1265, which equals 0.1273; the fifth is 0.1841 divided by 2.1265, which equals 0.0866; and the sixth is 0.0240 divided by 2.1265, which equals 0.0113. The sum of these six fusion weights is 1.0.

[0042] The process of obtaining the weighted predicted quality value is as follows: Six reference batches of finished product anti-HPV activity test values ​​are retrieved. These values ​​are anti-HPV activity indicators determined by in vitro cell biology experiments, expressed in units per milliliter. For example, the six test values ​​are 98, 102, 94, 89, 85, and 77. The fusion weight of each reference batch is multiplied by its test value: the first 98 multiplied by 0.3123 equals 30.61, the second 102 multiplied by 0.2781 equals 28.37, the third 94 multiplied by 0.1845 equals 17.34, the fourth 89 multiplied by 0.1273 equals 11.33, the fifth 85 multiplied by 0.0866 equals 7.36, and the sixth 77 multiplied by 0.0113 equals 0.87. Adding the weighted values ​​of the six batches together, 30.61 plus 28.37 plus 17.34 plus 11.33 plus 7.36 plus 0.87 equals 95.88. This summation of 95.88 is the predicted quality value of the gel production for the target production batch.

[0043] In another preferred embodiment of the present invention, the specific process of combining the quality prediction value with the instruction order number and the production instruction issuance time in step S6 and storing it into the historical batch database is as follows: The instruction order number for the target batch is "HPV20240513003". The product code field "HPV", the production date field "20240513", and the batch sequence number field "003" are extracted from this. These three fields are combined with the quality prediction value of 95.88 calculated in step S5 and the production instruction issuance time "2024-05-13 08:30:00" to form a fixed-length record entry. The structure of this fixed-length record entry is as follows: the product code field occupies 3 character positions, the production date field occupies 8 character positions, the batch sequence number field occupies 3 character positions, the quality prediction value occupies 6 character positions including the decimal point, and the production instruction issuance time occupies 19 character positions. All fields are concatenated sequentially with a fixed width.

[0044] The release control threshold range corresponding to the product code field "HPV" is retrieved from the preset release control parameter table. The release control parameter table stores the quality acceptance limits for each product specification. For product code "HPV", the lower limit of the release control threshold range is 80, and the upper limit is 120. The predicted quality value 95.88 is compared with this threshold range. 95.88 is greater than the lower limit 80 and less than the upper limit 120, falling within the release control threshold range. Therefore, a pending release identifier is generated, which is a predefined string value, such as "RELEASE". If the predicted quality value does not fall within this range, for example, below 80 or above 120, a pending review identifier is generated, which is another predefined string value, such as "REVIEW". In this embodiment, "RELEASE" is generated.

[0045] Record the current time of prediction generation in the format of year, month, day, hour, minute, and second, such as "20240513150000". Combine the aforementioned fixed-length record entry, the release flag "RELEASE", and the prediction generation time "20240513150000" into a complete prediction record. The entire content of this prediction record includes the product code HPV, production date 20240513, batch number 003, predicted quality value 95.88, production instruction issuance time 2024-05-13 08:30:00, release flag RELEASE, and prediction generation time 20240513150000. Store this prediction record in the historical batch database. The historical batch database is a persistent storage device that already stores the production process records, batch process characteristic sequences, and finished product anti-HPV activity test values ​​for all historical batches of this manufacturer.

[0046] Simultaneously, the instruction order number and corresponding process characteristic deviation of each reference batch in the reference batch set are compiled into list entries. Each list entry contains two fields: the reference batch instruction order number and the process characteristic deviation. Six reference batches generate six entries; for example, the first entry is "HPV20240510001, 0.34", the second is "HPV20240508002, 0.41", and so on. All six list entries are combined into a single list in ascending order of process characteristic deviation. This list, along with the previously generated prediction records, is sent via network communication to the production batch release review node. The release review node is the workstation responsible for batch release decisions in the production management process. Upon receiving the prediction records and the reference batch list, the reviewers can understand the prediction quality level of the target batch and the historical references upon which the prediction is based, and perform corresponding release or review operations based on the pending release or pending review flags.

[0047] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for predicting the production quality of alginate-sodium hyaluronate anti-HPV gel, characterized in that, Includes the following steps: S1. Input the instruction order number of the target production batch, which is associated with the batch number of alginate raw material, sodium hyaluronate raw material, crosslinking agent, reactor identification, production instruction issuance time, and production process record; S2. Obtain the operation records of the stirring shaft rotating forward and backward sequentially during the crosslinking stirring stage from the production process records, and form forward rotation load sequence and reverse rotation load sequence respectively; S3. The switching time from forward to reverse rotation is taken as the first switching point. A first number of load values ​​are intercepted in the reverse direction of the forward load sequence as the first leading sequence. A second number of load values ​​are intercepted in the forward direction of the reverse load sequence starting from the first switching point as the first following sequence. The average difference between adjacent load values ​​is calculated using the load values ​​at the end of the first leading sequence. The average difference is divided by the sampling interval, and the result is taken as the load change rate from forward to reverse rotation. The average load value of the stable segment far from the switching area in the reverse load sequence is superimposed with a preset fluctuation tolerance to form a steady-state recovery interval. The sampling time when the first load value falls into the steady-state recovery interval is located in the first following sequence. The time span between the sampling time and the first switching point is calculated as the recovery time from forward to reverse rotation. The switching time from reverse to forward rotation is taken as the second switching point, and the load change rate and recovery time from reverse to forward rotation are calculated in the same way. Obtain the baseline load change rate and baseline recovery time associated with the product specification code in the instruction sheet number. Divide the load change rate of each forward to reverse switch by the baseline load change rate, and the resulting ratio is used as the first change rate ratio. Divide the recovery time of the forward to reverse switch by the baseline recovery time, and the resulting ratio is used as the first recovery time ratio. Combine the first change rate ratio and the first recovery time ratio to form the switching process response characteristics of the forward to reverse switch. Convert the load change rate and recovery time of each reverse to forward switch into the second change rate ratio and the second recovery time ratio in the same way, and combine them to form the switching process response characteristics of the reverse to forward switch. Arrange the response characteristics of all switching processes in the crosslinking stirring stage in chronological order of the switching occurrence to form a batch process characteristic sequence. S4. Retrieve the process feature sequences of historical batches from the same manufacturer from the historical batch database, calculate the deviation of the process features between the target batch and each historical batch, and select a set of reference batches based on the deviation of the process features. The specific process for calculating the deviation of the process features is as follows: The batch process feature sequence of the target batch is recorded as the target sequence, and the batch process feature sequence of the historical batch is recorded as the comparison sequence. The target sequence is subjected to segmentation point detection. The switching point where the absolute value of the difference between the first change rate ratios of the response features of adjacent switching processes exceeds the preset change rate segmentation limit is marked as the segmentation point. The target sequence is divided into several target sequence segments using the segmentation points. According to the position of the segmentation point in the target sequence, the comparison sequence is divided into several comparison sequence segments at the same position in the comparison sequence. Align each target sequence segment with the corresponding comparison sequence segment position by position, calculate the sum of the squared difference of the first rate of change ratio and the squared difference of the first recovery time ratio at the aligned positions, use the sum as the single-point difference value of the position, accumulate all single-point difference values ​​between all target sequence segments and the corresponding comparison sequence segments, and use the accumulated result as the process characteristic deviation degree between the target batch and the historical batch. S5. Obtain the anti-HPV activity test value and production date of the reference batch of finished products. Determine the fusion weight by the process characteristic deviation and the interval between the production date of the reference batch and the production instruction of the target batch. Calculate the quality prediction value by weighting. S6. Combine the quality prediction value with the instruction sheet number and the production instruction issuance time to form a prediction record and store it in the historical batch database. Send the reference batch instruction sheet number and quality prediction value to the release review node.

2. The method for predicting the production quality of alginate-sodium hyaluronate anti-HPV gel according to claim 1, characterized in that, In S2, the specific processes for forming the forward load sequence and the reverse load sequence are as follows: The product specification code in the instruction sheet number is analyzed, and the forward rotation duration, reverse rotation duration, and direction pause duration are retrieved from the mixing cycle configuration table based on the product specification code. The cross-linking mixing stage consists of alternating forward rotation, pause, and reverse rotation duration segments. During the pause segment, the mixing shaft drive input is zero. Within each forward rotation duration segment, the load value of the drive mechanism is recorded at a fixed acquisition interval. The load values ​​of all forward rotation duration segments are arranged sequentially by acquisition time to form a forward rotation load sequence. Within each reverse rotation duration segment, the load value of the drive mechanism is recorded at the same fixed acquisition interval. The load values ​​of all reverse rotation duration segments are arranged sequentially by acquisition time to form a reverse rotation load sequence.

3. The method for predicting the production quality of alginate-sodium hyaluronate anti-HPV gel according to claim 1, characterized in that, In step S4, the specific process of selecting the reference batch set based on the process characteristic deviation is as follows: All historical batches participating in the comparison are sorted in ascending order of process feature deviation value to form a deviation sequence. The difference in process feature deviation between adjacent positions in the deviation sequence is calculated, and the larger process feature deviation in the group of adjacent positions with the largest difference is taken as the critical deviation. All historical batches in the deviation sequence whose process feature deviation is less than the critical deviation are extracted as a reference batch set.

4. The method for predicting the production quality of alginate-sodium hyaluronate anti-HPV gel according to claim 1, characterized in that, In step S5, the specific process of determining the fusion weight based on the process characteristic deviation and the interval between the reference batch production date and the target batch production instruction issuance time is as follows: For each reference batch in the reference batch set, a similarity score is calculated using the negative value of the process characteristic deviation of the reference batch as the exponent, with the natural constant as the base. The number of days between the production date of the reference batch and the production instruction time of the target batch is divided by a preset half-life to obtain the decay index. A time decay factor is calculated using the decay index as the exponent, with 0.5 as the base. The similarity score of the reference batch is multiplied by the time decay factor to obtain the fusion coefficient. The sum of the fusion coefficients of all reference batches in the reference batch set is calculated. The fusion coefficient of each reference batch is divided by the sum of the fusion coefficients to obtain the fusion weight of the reference batch. The process of weighted calculation to obtain the quality prediction value is as follows: the fusion weight of each reference batch is multiplied by the finished anti-HPV activity test value of the reference batch to obtain the single-batch weighted value. The single-batch weighted values ​​of all reference batches in the reference batch set are added together, and the sum is used as the quality prediction value.

5. The method for predicting the production quality of an alginate-sodium hyaluronate anti-HPV gel according to claim 1, characterized in that, In step S6, the specific process of combining the quality prediction value with the instruction order number and the production instruction issuance time to form a prediction record and store it in the historical batch database is as follows: Extract the product code field, production date field, and batch sequence field from the instruction order number, and combine them with the quality prediction value and the production instruction issuance time to form a fixed-length record entry. Retrieve the release control threshold range corresponding to the product code field, and compare the quality prediction value with the release control threshold range. If the quality prediction value falls within the release control threshold range, a pending release label is generated; if the quality prediction value falls outside the release control threshold range, a pending review label is generated. Combine the fixed-length record entry, label, and prediction generation time to form a prediction record, and store the prediction record in the historical batch database. The instruction sheet number and process characteristic deviation of each reference batch in the reference batch set are arranged into list items. All list items are combined into a list, and the list and prediction records are sent to the release review node.

Images