Drug curative effect evaluation method and system using industrial big data
By using industrial big data analysis, combined with self-supervised window coding and Bayesian fusion technology, the accuracy problem of cross-process cumulative effects in drug efficacy evaluation was solved, enabling reliable evaluation of drug efficacy and identification of causal paths.
Patent Information
- Application Number
- CN202511845056.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-09
- Publication Date
- 2026-03-10
AI Technical Summary
Existing methods for evaluating drug efficacy lack a systematic analysis of the cumulative effects across processes, fail to accurately understand the indirect impact of preceding processes on subsequent processes and the final drug efficacy, and do not consider the dynamic changes of equipment residues between different processes, resulting in insufficient accuracy of the evaluation results.
By combining industrial big data analysis with joint modeling of equipment process data and residual data, and employing self-supervised window coding, adjacency propagation mechanism, memory decay function and Bayesian fusion technology, a causal relationship graph is generated to achieve accurate evaluation of drug efficacy.
It improves the accuracy and reliability of drug efficacy assessment, can accurately capture the dynamic transfer patterns of equipment residues between different processes, identify complex causal pathways, and provide more reliable causal pathway support.
Smart Images
Figure CN121639030A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of pharmaceutical industry data analysis technology, specifically to a method and system for evaluating drug efficacy using industrial big data. Background Technology
[0002] In recent years, with the development of industrial automation and intelligent manufacturing technologies, equipment operation data, sensor monitoring data, and process parameter information in industrial production processes have been accumulated on a large scale, forming rich industrial big data. This industrial big data not only covers the operating status of production equipment under different processes, but also includes information on substances that may remain in the production process. Using this large-scale, multi-source, and multi-dimensional industrial data to quantitatively evaluate the impact on drug efficacy in the drug production process has become an important technical research direction in the pharmaceutical industry and data analysis field.
[0003] In existing technologies, current drug efficacy evaluation schemes are usually based on process optimization methods using process equipment data. For example, they analyze parameters such as equipment temperature, pressure, and stirring rate, or assess possible changes in drug activity through residue detection. The advantages of these methods are that they are relatively simple to operate, the amount of data is controllable, and they can provide a certain reference for evaluating changes in drug efficacy. However, these methods analyze at the single-batch level when evaluating efficacy, and lack precise quantitative modeling of the cumulative contribution across processes. This makes it difficult to accurately understand the indirect impact of preceding processes on subsequent processes and the final drug efficacy. At the same time, since different processes may have different residues in the equipment, it further exacerbates the incomplete understanding of the cumulative effect across processes, resulting in insufficient accuracy of the final drug efficacy evaluation results. Summary of the Invention
[0004] The purpose of this invention is to provide a method and system for evaluating drug efficacy using industrial big data, so as to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the technical solution of the present invention is as follows:
[0006] In a first aspect, the present invention discloses a method for evaluating drug efficacy using industrial big data, comprising the following steps:
[0007] Obtain the equipment process data and equipment residual data of the target object;
[0008] The equipment process data is subjected to dual-clock fault-tolerant synchronization processing to generate synchronized timing data, and the quality of the synchronized timing data is evaluated.
[0009] The synchronous time-series data is sliced into process segments and process feature vectors are generated based on self-supervised window encoding. The process feature vectors are then processed by weight mapping and adjacency propagation to generate a cumulative contribution matrix. The weights in the self-supervised window encoding process are set according to the quality evaluation results of the corresponding process segments.
[0010] The residual data of the device is divided into sliding windows, and the residual data of the device in each sliding window is cumulatively calculated using a pre-built timestamp-based memory decay function;
[0011] A windowed joint retrieval is performed on the synchronous time-series data and the cumulative calculation results to generate candidate causal event pairs. The cumulative contribution matrix is used as a prior constraint to fit the cumulative contribution of the candidate causal events and generate a causal relationship graph.
[0012] The cumulative contribution matrix, cumulative calculation results, and causal relationship graph are fused using Bayesian methods to generate drug efficacy evaluation results.
[0013] Secondly, this invention discloses a drug efficacy evaluation system utilizing industrial big data, comprising:
[0014] The data acquisition module is used to acquire equipment process data and equipment residual data of the target object;
[0015] The quality assessment module is used to perform dual-clock fault-tolerant synchronization processing on the equipment process data, generate synchronized timing data, and perform quality assessment on the synchronized timing data.
[0016] The process contribution calculation module is used to slice the synchronous time-series data into process segments and generate process feature vectors based on self-supervised window encoding. The process feature vectors are then processed by weight mapping and adjacency propagation to generate a cumulative contribution matrix.
[0017] The device residual accumulation module is used to divide the device residual data into sliding windows and perform cumulative calculations on the device residual data in each sliding window using a pre-built timestamp-based memory decay function.
[0018] The causal path identification module is used to perform windowed joint retrieval on the synchronous time-series data and cumulative calculation results, generate candidate causal event pairs, and use the cumulative contribution matrix as a prior constraint to perform cumulative contribution fitting on the candidate causal events to generate a causal relationship graph.
[0019] The efficacy evaluation module is used to perform Bayesian fusion on the cumulative contribution matrix, cumulative calculation results and the causal relationship graph to generate drug efficacy evaluation results.
[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0021] 1. This scheme establishes a linkage mechanism between the sensitivity table and the transfer coefficient, enabling the process contribution matrix to reflect the dynamic changes of residual effects in real time. It can accurately capture the dynamic transfer patterns of equipment residues between different processes. At the same time, confidence correction eliminates the interference of low-quality data on the calculation process, thereby improving the accuracy of residual effect analysis in the drug efficacy evaluation process.
[0022] 2. This scheme, by introducing a time-delay regression model and a dynamic latent variable alignment mechanism, combined with the prior constraints of the cumulative contribution matrix, can more accurately identify complex causal paths involving time lag and residue transmission, avoiding misjudgments caused by data asynchrony or the failure to quantify cross-process influences, thus providing more reliable causal path support for drug efficacy evaluation. Attached Figure Description
[0023] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:
[0024] Figure 1 This is a flowchart illustrating the steps of the drug efficacy evaluation method utilizing industrial big data of the present invention.
[0025] Figure 2 A schematic diagram of the process for generating the updated cumulative contribution matrix provided by the present invention;
[0026] Figure 3 A schematic diagram of the process for generating a comprehensive efficacy score provided by the present invention;
[0027] Figure 4 This is a schematic diagram of the process for generating drug efficacy evaluation results provided by the present invention;
[0028] Figure 5 This is a schematic diagram of the module functions of the drug efficacy evaluation system utilizing industrial big data provided by the present invention. Detailed Implementation
[0029] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.
[0030] Application Overview:
[0031] In existing technologies, drug efficacy assessment mainly relies on equipment parameters and residue detection data of a single batch, lacking a systematic analysis of the cumulative effect across processes. Existing methods are unable to quantify the indirect impact of preceding processes on subsequent processes and the final drug efficacy, and do not consider the dynamic changes of equipment residues between different processes, leading to biases in the assessment results.
[0032] To address the aforementioned issues, this study found that the core deficiency of existing solutions lies in the lack of a dynamic correlation model between processes and the failure to integrate the temporal cumulative effect of equipment residues. By analyzing the characteristics of industrial big data, this study proposes to jointly model equipment process data and residue data, and to improve data reliability by utilizing temporal synchronization and quality assessment. Furthermore, to address the difficulty in quantifying cross-process impacts, a self-supervised window encoding and adjacency propagation mechanism is designed, combined with a memory decay function to capture the cumulative effect of residues. Finally, Bayesian fusion is used to achieve collaborative analysis of multi-source data, thereby constructing a complete causal correlation assessment system.
[0033] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0034] Example 1:
[0035] Please see Figure 1 The drug efficacy evaluation method utilizing industrial big data includes the following steps:
[0036] Obtain the equipment process data and equipment residual data of the target object;
[0037] Perform dual-clock fault-tolerant synchronization processing on the equipment process data to generate synchronized timing data, and conduct quality assessment on the synchronized timing data;
[0038] Synchronous time-series data is sliced into process segments and process feature vectors are generated based on self-supervised window encoding. The process feature vectors are then processed by weight mapping and adjacency propagation to generate a cumulative contribution matrix. The weights in the self-supervised window encoding process are set according to the quality evaluation results of the corresponding process segments.
[0039] The residual data of the equipment is divided into sliding windows, and the residual data of the equipment in each sliding window is accumulated by performing a pre-built memory decay function based on timestamps.
[0040] A windowed joint retrieval is performed on the synchronous time-series data and the cumulative calculation results to generate candidate causal event pairs. Then, with the cumulative contribution matrix as a prior constraint, the cumulative contribution of the candidate causal events is fitted to generate a causal relationship graph.
[0041] Bayesian fusion of the cumulative contribution matrix, cumulative calculation results, and causal relationship diagram is performed to generate drug efficacy evaluation results.
[0042] Among them, dual-clock fault-tolerant synchronization processing refers to correcting the timestamp deviation that may exist in the equipment process data. Specifically, it can be achieved by using a timestamp alignment algorithm combined with the sliding window averaging method.
[0043] Self-supervised window encoding refers to feature extraction from process slice data, which can be achieved by combining convolutional neural networks with attention mechanisms.
[0044] Adjacency transitive processing refers to the information transmission of process feature vectors between processes, which can be implemented using the message passing mechanism in graph neural networks.
[0045] The memory decay function refers to the time-weighted accumulation of residual data in the device, which can be implemented by combining an exponential decay model with a sliding window integral.
[0046] Bayesian fusion refers to the integration of multi-source data using a probabilistic model, which can be achieved by combining Markov chain Monte Carlo sampling with posterior distribution analysis.
[0047] This solution achieves dynamic weight adjustment through self-supervised window coding and quality assessment, enhancing the objectivity of data-driven approaches. Bayesian fusion balances the confidence levels of multi-source data, strengthening the robustness of assessment results. Through these technical solutions, this application can accurately quantify the impact of cross-process cumulative effects on drug efficacy, addressing assessment bias caused by dynamic changes in equipment residues. Self-supervised coding and adjacency propagation mechanisms effectively capture nonlinear correlations between processes; memory decay functions and windowing-based joint retrieval enable temporal modeling of cumulative residue effects; and Bayesian fusion integrates multi-dimensional data, improving the credibility of assessment results. This method provides a reliable basis for efficacy evaluation in continuous production batches, demonstrating significant advantages, especially in handling complex process interaction scenarios.
[0048] The following details the process of quality assessment for synchronized time-series data, which also includes generating a sampling qualification report, specifically:
[0049] Batch alignment and standardization are performed on the equipment residual data to generate standard residual data. The standard residual data is then subjected to internal and external consistency checks based on the batch to generate a sampling qualification report.
[0050] Among them, batch alignment and standardization refers to matching and aligning the residual equipment data of different production batches according to the time series and eliminating the dimensional differences between different batches. Specifically, it can be achieved by using a timestamp matching algorithm combined with a data normalization method.
[0051] Internal and external consistency testing refers to verifying the consistency of residual data within the same batch and the differences in residual data between different batches through statistical testing methods. Specifically, it can be achieved by combining homogeneity of variance testing with cluster analysis.
[0052] This solution achieves time synchronization of cross-batch data through dynamic time warping, and verifies the internal consistency and external differences of batch data through both homogeneity of variance test and cluster analysis. This effectively solves the problem of decreased reliability of residual data caused by equipment operating cycle fluctuations and sensor drift. Through the above technical solution, this application can ensure the comparability of equipment residual data in both time and numerical dimensions, and at the same time verify the quality and reliability of batch data from a statistical perspective. This provides a rigorously verified data foundation for subsequent cumulative contribution matrix updates and efficacy evaluation, and avoids the accumulation of efficacy evaluation errors caused by residual data quality defects.
[0053] The following details the process of accumulating the device residual data within each sliding window using a pre-built timestamp-based memory decay function, including updating the cumulative contribution matrix, specifically:
[0054] Using process feature vectors as independent variables, the cumulative calculation results are subjected to regression decomposition and regularization with confidence correction based on quality assessment results to generate a process residual sensitivity table.
[0055] The transmission coefficients in the adjacency transmission process are corrected based on the process residual sensitivity table, and an updated cumulative contribution matrix is generated.
[0056] Among them, the process residue sensitivity table refers to a quantitative indicator that reflects the degree of correlation between different process characteristics and the cumulative effect of residue data. Specifically, it can be implemented using a partial least squares regression method with confidence weighting for quality assessment.
[0057] The transmission coefficient refers to the weight allocation parameter among nodes in the adjacency transmission process of the process feature vector. Specifically, it can be implemented by a dynamic adjustment mechanism based on the sensitivity table values. By mapping the correlation strength of each process in the sensitivity table to the weight ratio of the transmission path, a dynamic correlation between residual effect and process contribution is established.
[0058] This solution establishes a linkage mechanism between the sensitivity table and the transfer coefficient, enabling the process contribution matrix to reflect the dynamic changes of residual effects in real time. Simultaneously, confidence correction eliminates the interference of low-quality data on the calculation process, effectively improving the reliability of the cumulative contribution matrix. Through this technical solution, this application can accurately capture the dynamic transfer patterns of equipment residues between different processes, solving the problem of modeling deviations in cross-process cumulative effects caused by fixed transfer coefficients in existing technologies. By correcting the sensitivity calculation using quality assessment results, the impact of low-quality process data on the overall assessment can be effectively suppressed, thereby improving the accuracy of residual effect analysis in drug efficacy evaluation.
[0059] The following describes how to fit the cumulative contribution matrix to candidate causal event pairs using the cumulative contribution matrix as a prior constraint, and generate a causal relationship graph, specifically including:
[0060] Based on synchronous time-series data, time-delay regression fitting is performed on candidate causal event pairs to generate a causal relationship diagram;
[0061] Furthermore, during the time-delay regression fitting process, the updated cumulative contribution matrix is used as a prior constraint, and the cumulative calculation results are used as the time-series alignment input of the latent variables.
[0062] Among them, time delay regression fitting refers to inferring causal relationships by establishing a time lag correlation model between candidate events. Specifically, it can be achieved by combining the dynamic time warping algorithm with the Granger causality test method.
[0063] The cumulative contribution matrix as a prior constraint refers to using the matrix obtained by weighting and adjacency transiting the process feature vectors as the prior distribution parameters in the Bayesian framework. Specifically, the Gibbs sampling method can be used to transform the matrix elements into probability distribution parameters.
[0064] The time-aligned input of latent variables refers to embedding the cumulative calculation results of equipment residual data as latent variables into the regression model. Specifically, the sliding window mean and the weighted value of the memory decay function can be used as the input feature vector.
[0065] This solution, by introducing a time-delay regression model and a dynamic latent variable alignment mechanism, combined with the prior constraint of the cumulative contribution matrix, can more accurately identify complex causal paths involving time lag and residue transmission, avoiding misjudgments caused by data asynchrony or the lack of quantification of cross-process influences. Through the above technical solution, this application can effectively solve the problems of time-series asynchrony and omission of residue influences in the identification of cross-process causal paths by traditional methods, significantly improving the accuracy and interpretability of causal relationship diagrams. By using the cumulative contribution matrix as a prior constraint, spurious associations between low-correlation processes can be suppressed. By using the cumulative results of residue data as the latent variable input for time-series alignment, the transmission effect of residues between processes can be dynamically captured, thereby providing more reliable causal path support for drug efficacy evaluation.
[0066] The following describes the Bayesian fusion of the cumulative contribution matrix, cumulative calculation results, and causal relationship graph to generate drug efficacy evaluation results, including:
[0067] Hierarchical confidence weighted fusion of the cumulative contribution matrix, cumulative calculation results and causal relationship diagram is performed to generate a comprehensive efficacy score;
[0068] The expected loss of efficacy was obtained by weighting the comprehensive efficacy score and causal relationship diagram by production batch.
[0069] The overall efficacy score and expected loss of drug efficacy are combined in a structured manner to generate drug efficacy assessment results.
[0070] Among them, hierarchical confidence weighted fusion refers to multi-dimensional weight allocation based on the reliability differences of data sources. Specifically, it can be achieved by using a Bayesian network model to jointly model the cumulative contribution matrix as the process contribution observation item, the cumulative calculation result as the long-term risk correction item, and the causal relationship graph as the structural constraint. The confidence level can be dynamically adjusted through the conditional probability relationship between different data sources.
[0071] Weighted calculation refers to differentiated processing based on the degree of impact of different production batches, which can be achieved by mapping the risk factor weight vector to the batch weight coefficient.
[0072] Structured combination refers to integrating multidimensional assessment indicators into a unified output format. Specifically, decision tree classification or matrix splicing can be used to combine the comprehensive efficacy score with the maximum expected benefit matching results.
[0073] This solution achieves dynamic weight allocation of multi-source heterogeneous data through a Bayesian fusion mechanism. For example, it jointly analyzes the long-term cumulative effects of equipment residues with causal paths between processes, overcoming the evaluation bias caused by data isolation in traditional methods. Simultaneously, the structured combination method can effectively distinguish the differences in the impact of different production batches on the final drug efficacy, for example, by capturing the temporal fluctuation characteristics of equipment operating status through risk factor weight vectors. Through the above technical solutions, this application can accurately quantify the degree of influence of cross-process cumulative effects on drug efficacy, for example, by identifying the interaction patterns between key processes through hierarchical confidence fusion. Furthermore, the expected drug efficacy loss assessment based on dynamic weight calculation can effectively reflect the quality fluctuations of different production batches, for example, by locating abnormal production links through the correlation analysis of batch characteristics and causal paths. The final structured evaluation results can provide multi-dimensional decision-making basis for production process optimization, for example, guiding equipment parameter adjustment schemes based on expected benefit ranking.
[0074] The following describes the hierarchical confidence-weighted fusion of the cumulative contribution matrix, cumulative calculation results, and causal relationship diagram to generate a comprehensive efficacy score, including:
[0075] An iterative sampling process is performed on the pre-constructed data structure until the preset number of iterations is reached, generating a comprehensive efficacy score;
[0076] The pre-constructed data structure is built by using the cumulative contribution matrix as the process contribution observation, the cumulative calculation result as the long-term risk correction term, the causal relationship diagram as the structural constraint, and the quality assessment result and the sampling qualification report as the prior weight factor.
[0077] The iterative sampling process refers to optimizing parameter estimation by repeatedly sampling data. Specifically, it can be implemented using the Markov chain Monte Carlo sampling method, which generates posterior samples for statistical analysis to approximate the true distribution.
[0078] Hierarchical confidence-weighted fusion refers to the hierarchical quantitative integration of the reliability of different data sources. Specifically, it can be achieved using the weight allocation method under the Bayesian inference framework, by introducing quality assessment results and sampling qualification reports to adjust the confidence of different data sources.
[0079] Prior weighting factors refer to parameter adjustment coefficients pre-set based on data quality assessment results. Specifically, they can be constructed using normalized batch consistency test scores and quality assessment scores, and are used to balance the contribution ratios of different data sources during the fusion process.
[0080] This approach constructs a multi-dimensional data structure encompassing process contributions, long-term risks, and structural constraints. Combined with iterative sampling and confidence-weighted fusion mechanisms, it effectively integrates the temporal correlations of indirect influences and residual effects between different processes, addressing the incomplete modeling of cross-process cumulative effects in traditional methods. Through this technical solution, this application achieves deep fusion of multi-source heterogeneous data in drug efficacy evaluation. By introducing joint modeling of process contribution observations and long-term risk correction terms, coupled with a dynamic weight adjustment mechanism, it significantly enhances the comprehensive quantitative capability of cross-process cumulative effects and residual risks. This method effectively avoids evaluation biases caused by single-batch analysis, providing a more comprehensive quantitative basis for efficacy changes during drug production.
[0081] The following describes the specific steps involved in performing iterative sampling on a pre-built data structure:
[0082] In each iteration of sampling, Markov chain Monte Carlo sampling is performed on the pre-constructed data structure to generate posterior samples. Statistical analysis is then performed on the posterior samples to generate risk factor weight vectors. Based on the risk factor weight vectors, a weighted average is calculated on the cumulative contribution matrix and the cumulative calculation results to obtain the effect value.
[0083] The parameters of the causal relationship graph are updated based on the risk factor weight vector until the preset number of iterations is reached. The risk factor weight vector corresponding to the current iteration number is output, and the effect values in each iteration sampling process are weighted and averaged to obtain the comprehensive efficacy score.
[0084] Markov chain Monte Carlo sampling refers to a random sampling method based on probability distribution;
[0085] The risk factor weight vector refers to the weight parameters that reflect the contribution and residual impact of different processes. Specifically, it can be realized through Bayesian regression coefficients or principal component analysis loading coefficients.
[0086] The effect value is a comprehensive index of process contribution and residual impact obtained through weighted calculation, which can be achieved by linear weighting or nonlinear fusion methods.
[0087] This scheme, through iterative sampling and parameter collaborative update mechanisms, can adaptively adjust the contribution weights of different processes. Simultaneously, by combining the structural constraints of the causal relationship graph, it effectively captures indirect influence paths across processes, significantly improving the accuracy of efficacy evaluation in complex production environments. Through the above technical solutions, this application solves the problem of evaluation result bias caused by fixed weight allocation in traditional methods. By dynamically adjusting weights and co-optimizing parameters, it improves the accuracy of modeling cumulative effects across processes. Furthermore, by combining Markov chain Monte Carlo sampling with Bayesian statistics, it effectively reduces the impact of high-dimensional data noise on the evaluation results, enhancing the model's adaptability to complex industrial data. Therefore, in scenarios with equipment residue interference and multi-process interactions, it can generate more reliable drug efficacy evaluation results, avoiding the risk of misjudgment due to local optima.
[0088] The following describes the weighted calculation of the comprehensive efficacy score and causal relationship diagram by production batch to obtain the expected loss of efficacy, including:
[0089] The expected loss of efficacy was obtained by weighting the comprehensive efficacy score and causal relationship diagram by production batch.
[0090] The weight coefficients in the weighted calculation process are obtained by mapping the risk factor weight vector corresponding to the current iteration number.
[0091] Among them, weighted calculation refers to assigning differentiated weights to data from different production batches to reflect their degree of influence on drug efficacy loss, which can be achieved by linear combination or nonlinear transformation.
[0092] The risk factor weight vector refers to a set of multidimensional parameters generated through an iterative sampling process that reflects the contribution of each process, residual risk, and the strength of causal relationships. Specifically, it can be implemented by combining Markov chain Monte Carlo sampling with Bayesian inference.
[0093] This solution combines iteratively optimized risk factors with production batch characteristics through a dynamic weight mapping mechanism, enabling the calculation of efficacy loss to adaptively reflect the complex relationships across processes and batches, thereby improving the accuracy of the evaluation results. Through the above technical solution, this application solves the problem of efficacy loss assessment bias caused by ignoring differences in production batches in the prior art. By using dynamic weight coefficients, it achieves differentiated modeling of process data for different batches, enabling the calculation of expected efficacy loss to more accurately reflect the long-term impact of residue accumulation and process parameter fluctuations in the actual production process, providing reliable data support for the optimization of drug production processes.
[0094] The following describes the structured combination of comprehensive efficacy score and expected loss of efficacy to generate drug efficacy assessment results, specifically including:
[0095] The causal relationship diagram and the process residue sensitivity table are matched with a predefined set of action candidates. For each matching result, the cumulative contribution matrix and the change value of the cumulative calculation result are calculated respectively when the matching result is executed to obtain the expected benefits.
[0096] The matching results corresponding to the comprehensive efficacy score, expected loss of drug efficacy, and maximum expected benefit are structured and combined to generate drug efficacy evaluation results.
[0097] Among them, the causal relationship graph refers to a graphical data structure generated by fitting the cumulative contribution of candidate causal event pairs. Specifically, it can be implemented by using a time-delay regression algorithm combined with the adjacency transit coefficient update.
[0098] The process residue sensitivity table is a quantitative index that reflects the sensitivity of the process feature vector to the calculation results of equipment residue accumulation, generated by regression decomposition. Specifically, it can be implemented by using a linear regression model with confidence correction combined with regularization.
[0099] The action candidate set refers to a predefined set of equipment operation adjustment strategies, which can be constructed by combining historical process optimization records with expert experience rules.
[0100] Expected benefits refer to the changes in the cumulative contribution matrix and cumulative calculation results after performing a specific action candidate. Specifically, they can be calculated by combining Monte Carlo simulation with Bayesian inference.
[0101] This approach, by introducing a matching mechanism for candidate action sets, can dynamically simulate the impact path of different process parameter adjustments on the cumulative effects across processes. Compared to static analysis relying solely on historical data, this method establishes an operable closed-loop feedback mechanism for process optimization and efficacy evaluation, directly linking efficacy evaluation results to specific production parameter adjustment schemes. Through this technical solution, this application effectively addresses the problem of traditional drug efficacy evaluation methods lacking the ability to simulate process adjustment strategies. By establishing a matching mechanism for candidate action sets and a mechanism for calculating expected benefits, the impact of different process parameter adjustments on the cumulative effects across processes can be accurately predicted, thereby generating efficacy evaluation results containing executable optimization suggestions. This method significantly enhances the guiding value of evaluation results for production practice, transforming efficacy evaluation from a simple result judgment into a decision support system that includes improvement paths.
[0102] Example 2:
[0103] Please see Figure 5 Drug efficacy evaluation systems utilizing industrial big data include:
[0104] The data acquisition module is used to acquire equipment process data and equipment residual data of the target object;
[0105] The quality assessment module is used to perform dual-clock fault-tolerant synchronization processing on equipment process data, generate synchronized timing data, and perform quality assessment on the synchronized timing data.
[0106] The process contribution calculation module is used to slice synchronous time-series data into process segments and generate process feature vectors based on self-supervised window encoding. The process feature vectors are then processed by weight mapping and adjacency propagation to generate a cumulative contribution matrix.
[0107] The equipment residual accumulation module is used to divide the equipment residual data into sliding windows and perform cumulative calculations on the equipment residual data in each sliding window using a pre-built timestamp-based memory decay function.
[0108] The causal path identification module is used to perform windowed joint retrieval on synchronous time-series data and cumulative calculation results, generate candidate causal event pairs, and use the cumulative contribution matrix as a prior constraint to fit the cumulative contribution of the candidate causal events to generate a causal relationship graph.
[0109] The efficacy evaluation module is used to perform Bayesian fusion of the cumulative contribution matrix, cumulative calculation results, and causal relationship graph to generate drug efficacy evaluation results.
[0110] This embodiment has the same technical effects as Embodiment 1.
[0111] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. The data mentioned in this application, when used for calculations, have undergone normalization and other preprocessing to achieve dimensional uniformity.
[0112] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1.A method for evaluating drug efficacy using industrial big data, characterized by, The method comprises the following steps: obtaining device process data and device residual data of a target object; performing double-clock fault-tolerant synchronization processing on the device process data to generate synchronized timing data, and performing quality evaluation on the synchronized timing data; slicing the synchronized timing data into process slices and generating process feature vectors based on self-supervised window coding, and performing weight mapping and adjacency transfer processing on the process feature vectors to generate a cumulative contribution matrix; wherein the weights in the self-supervised window coding process are set according to the quality evaluation results of the corresponding process slices; dividing the device residual data into sliding windows, and performing cumulative calculation on the device residual data in each sliding window through a pre-constructed timestamp-based memory decay function; performing windowed joint retrieval on the synchronized timing data and the cumulative calculation results to generate candidate causal event pairs, and fitting the cumulative contribution of the candidate causal event pairs to the cumulative contribution matrix as a priori constraint to generate a causal correlation graph; performing Bayesian fusion on the cumulative contribution matrix, the cumulative calculation results and the causal correlation graph to generate a drug efficacy evaluation result. 2.The method for evaluating drug efficacy using industrial big data according to claim 1, characterized in that: After the quality evaluation of the synchronized timing data, a sampling qualified report is generated, specifically including: performing batch alignment and standardization processing on the device residual data to generate standard residual data, and performing internal and external consistency testing on the standard residual data based on batches to generate a sampling qualified report. 3.The method for evaluating drug efficacy using industrial big data according to claim 2, characterized in that: After performing cumulative calculation on the device residual data in each sliding window through a pre-constructed timestamp-based memory decay function, the cumulative contribution matrix is updated, specifically including: regression decomposition and regularization processing of the cumulative calculation results with confidence correction based on the quality evaluation results, generating a process residual sensitivity table; According to the process residual sensitivity table, the transmission coefficient in the adjacency transfer processing process is corrected to generate an updated cumulative contribution matrix. 4.The method for evaluating drug efficacy using industrial big data according to claim 3, characterized in that: Fitting the cumulative contribution of the candidate causal event pairs to the cumulative contribution matrix as a priori constraint to generate a causal correlation graph, specifically including: According to the synchronized timing data, the candidate causal event pairs are fitted by time delay regression to generate a causal correlation graph; And in the time delay regression fitting process, the updated cumulative contribution matrix is used as a priori constraint, and the cumulative calculation results are used as the time sequence alignment input of the latent variables. 5.The method for evaluating drug efficacy using industrial big data according to claim 3, characterized in that: The Bayesian fusion of the cumulative contribution matrix, the cumulative calculation results and the causal correlation graph to generate a drug efficacy evaluation result specifically includes: performing hierarchical confidence weighted fusion on the cumulative contribution matrix, the cumulative calculation results and the causal correlation graph to generate a comprehensive efficacy score; performing weighted calculation on the comprehensive efficacy score and the causal correlation graph according to the production batch to obtain an expected efficacy loss; performing structured combination on the comprehensive efficacy score and the expected efficacy loss to generate a drug efficacy evaluation result. 6.The method for evaluating drug efficacy using industrial big data according to claim 5, characterized in that: The hierarchical confidence weighted fusion of the cumulative contribution matrix, the cumulative calculation results and the causal correlation graph to generate a comprehensive efficacy score specifically includes: performing an iterative sampling process on the pre-constructed data structure until a preset number of iterations is reached to generate a comprehensive efficacy score; The pre-constructed data structure is constructed by taking the cumulative contribution matrix as a process contribution observation item, taking the cumulative calculation result as a long-term risk correction item, taking the causal association graph as a structural constraint, and taking the quality evaluation result and the sampling qualified report as a prior weight factor. 7.The method for evaluating drug efficacy using industrial big data according to claim 6, characterized in that: The iterative sampling process on the pre-constructed data structure specifically includes: In each iteration sampling process, Markov chain Monte Carlo sampling is performed on the pre-constructed data structure to generate posterior samples, and statistical analysis is performed on the posterior samples to generate a risk factor weight vector. The cumulative contribution matrix and the cumulative calculation result are weighted and averaged according to the risk factor weight vector to obtain an effect value. According to the risk factor weight vector, the causal association graph is updated until a preset number of iterations is reached, and the risk factor weight vector corresponding to the current iteration number is output. The effect values in each iteration sampling process are weighted and averaged to obtain a comprehensive efficacy score. 8.The method for evaluating drug efficacy using industrial big data according to claim 7, characterized in that: The comprehensive efficacy score and the causal association graph are weighted and calculated according to the production batch to obtain an expected efficacy loss, specifically including: The comprehensive efficacy score and the causal association graph are weighted and calculated according to the production batch to obtain an expected efficacy loss. The weight coefficient in the weighting calculation process is obtained by mapping the risk factor weight vector corresponding to the current iteration number. 9.The method for evaluating drug efficacy using industrial big data according to claim 5, characterized in that: The comprehensive efficacy score and the expected efficacy loss are structurally combined to generate a drug efficacy evaluation result, specifically including: The causal association graph and the process residual sensitivity table are matched with a pre-defined action candidate set, and for each matching result, the change value of the cumulative contribution matrix and the cumulative calculation result is calculated respectively under the condition of executing the matching result to obtain an expected benefit; The matching results corresponding to the comprehensive efficacy score, the expected efficacy loss and the maximum expected benefit are structurally combined to generate a drug efficacy evaluation result. 10.A system for evaluating drug efficacy using industrial big data, characterized by, It includes: A data acquisition module for acquiring device process data and device residual data of a target object; A quality evaluation module for performing double-clock fault-tolerant synchronization processing on the device process data to generate synchronized timing data, and performing quality evaluation on the synchronized timing data; A process contribution calculation module for slicing the synchronized timing data and generating a process feature vector based on a self-supervised window encoding, and performing weight mapping and adjacency transfer processing on the process feature vector to generate a cumulative contribution matrix; A device residual accumulation module for dividing the device residual data into sliding windows, and performing cumulative calculation on the device residual data in each sliding window by using a pre-constructed timestamp-based memory decay function; A causal path identification module for performing windowed joint retrieval on the synchronized timing data and the cumulative calculation result to generate candidate causal event pairs, and performing cumulative contribution fitting on the candidate causal events under the prior constraint of the cumulative contribution matrix to generate a causal association graph; A drug efficacy evaluation module is configured to perform Bayesian fusion on the accumulated contribution matrix, the accumulated calculation result, and the causal association graph to generate a drug efficacy evaluation result.