Industrial production process parameter optimization, regulation and control method and system based on knowledge graph

Abstract

The invention provides an industrial production process parameter optimization, regulation and control method and system based on a knowledge graph, and relates to the technical field of industrial production, and the method comprises the steps: collecting process parameter data, constructing a dynamic migration matrix, recognizing a process parameter combination through parameter clustering analysis, obtaining state transition characteristics, and carrying out dynamic evolution analysis to update the migration matrix. And calculating an optimal parameter regulation and control path and formulating an optimal regulation and control strategy. According to the invention, real-time optimization of the incidence relation of the process parameters can be realized, dynamic balance of optimization and regulation of the process parameters is achieved, and the industrial production efficiency and the product quality are improved.

Description

Technical Field

[0001] The present invention relates to industrial production technology, and in particular to a method and system for optimizing and controlling industrial production process parameters based on a knowledge graph. Background Art

[0002] In traditional industrial production, process parameter settings often rely on expert experience and pre-set fixed models, making them difficult to adapt to dynamic changes in the production process. Knowledge graphs, as a technical means of representing complex relationships, can effectively describe and analyze the correlations between process parameters, providing new ideas and methods for process parameter optimization.

[0003] At present, the main problems in the optimization and regulation of industrial production process parameters include: the complex correlation between process parameters is difficult to capture accurately, and existing methods are mostly based on static models, which cannot reflect the dynamic change characteristics of process parameters under different production conditions, resulting in limited parameter optimization effects; process parameter regulation lacks systematicness and foresight, and mostly adopts single parameter or local parameter optimization methods, which makes it difficult to achieve multi-parameter collaborative optimization and unable to achieve global optimality; the relationship between process parameter optimization and product quality is difficult to quantify and analyze, and existing technologies cannot achieve real-time dynamic optimization of parameters while ensuring product quality, resulting in a contradiction between parameter regulation and quality control, affecting production stability.

[0004] Knowledge graph technology offers new possibilities for solving these problems. By establishing semantic associations and reasoning capabilities between process parameters, we can more comprehensively understand the interaction mechanisms between them, providing more intelligent and precise technical support for process parameter optimization and regulation. Knowledge graph-based industrial production process parameter optimization and regulation methods can implement parameter association analysis, state transition prediction, and optimized path planning, effectively improving the level of intelligent industrial production. Summary of the Invention

[0005] The embodiments of the present invention provide a method and system for optimizing and controlling industrial production process parameters based on knowledge graphs, which can solve problems in the prior art.

[0006] A first aspect of an embodiment of the present invention provides a method for optimizing and controlling industrial production process parameters based on a knowledge graph, comprising: Collecting real-time process parameter data and historical process parameter data of industrial production equipment, analyzing the correlation characteristics and change patterns between process parameters based on the real-time process parameter data and the historical process parameter data, and constructing a process parameter dynamic migration matrix; Performing parameter cluster analysis on the process parameter dynamic migration matrix to identify process parameter combinations and obtain process parameter state transition characteristics; performing dynamic evolution analysis based on the process parameter state transition characteristics and updating the process parameter dynamic migration matrix to achieve real-time optimization of process parameter association relationships; Based on the process parameter dynamic migration matrix, combined with the current state and target state of the process parameters, the optimal parameter control path is calculated; according to the optimal parameter control path, the parameter optimization target is set, and the control effect of the process parameters and the changing trend of product quality are comprehensively considered to achieve a dynamic balance between process parameter optimization and control; A process parameter optimization and control strategy is calculated based on the real-time process parameter data and the process parameter dynamic migration matrix, and the process parameter optimization and control strategy is sent to industrial production equipment to execute process parameter adjustment.

[0007] Analyzing the correlation characteristics and change patterns between the process parameters based on the real-time process parameter data and the historical process parameter data, and constructing a process parameter dynamic migration matrix includes: Calculating a time-lag correlation coefficient between process parameters based on the real-time process parameter data and the historical process parameter data; The process parameters are grouped according to the values ​​of the time-lag correlation coefficients, the process parameters having the time-lag correlation coefficients greater than a preset time-lag threshold are divided into a first correlation group, and a process parameter state interval is constructed based on the process parameters in the strong correlation group; Based on the state interval of the process parameters, the state transition frequency of the process parameters at adjacent moments is counted, and the state transition probability of the process parameters is calculated according to the state transition frequency; the historical state transition probability and the real-time state transition probability of the process parameters are weightedly fused using the time-lag correlation coefficient to generate a dynamic migration matrix of the process parameters.

[0008] Performing parameter cluster analysis on the process parameter dynamic migration matrix, identifying process parameter combinations, and obtaining process parameter state transition characteristics includes: Constructing a state transition sequence similarity of process parameters, using the state transition sequence similarity as an initial clustering distance, and updating the initial clustering distance using a Lance-Williams parameter to obtain a clustering distance matrix of process parameters; calculating the intra-class closeness and inter-class separation of process parameter combinations based on the clustering distance matrix; Calculating an importance score of a process parameter combination according to the intra-class tightness and the inter-class separation, and screening a process parameter combination based on the importance score; performing stability verification on the process parameter combination to obtain a stability index of the process parameter combination; extract a main transition path of the process parameter combination based on the stability index, calculate a transition rate of the main transition path, and take the transition rate, the stability index, and the importance score as state transition features of the process parameter combination.

[0009] The initial clustering distance is updated using a Lance-Williams parameter to obtain a clustering distance matrix of the process parameters, including: The Lance-Williams parameter is constructed according to the number of samples in the process parameter category, and the Lance-Williams parameter is calculated based on the number of samples in two process parameter categories to be merged; The initial clustering distance is updated according to the Lance-Williams parameter, and the Lance-Williams parameter and the initial clustering distance are combined by weighting to obtain an updated distance after merging of the process parameter categories; The clustering distance matrix of the process parameters is generated according to the updated distance, and the clustering distance matrix records the distance relationship between the process parameter categories, and the symmetry of the clustering distance matrix is verified.

[0010] Based on the process parameter dynamic migration matrix, the current state and the target state of the process parameters are combined to calculate an optimal parameter control path, including: An evaluation function is constructed according to the transition probability in the process parameter dynamic migration matrix, the evaluation function reflects the transition difficulty between different process parameter states, transition probabilities greater than a preset migration threshold are identified from the process parameter dynamic migration matrix, process parameter states corresponding to the transition probabilities are determined to construct a reachable state set, and a state transition graph is constructed according to the state distribution of the reachable state set and the evaluation function; The state node intervals in the state transition graph are calibrated, the transition difficulty between adjacent state nodes is calculated based on the evaluation function, and the transition difficulty is calibrated between the corresponding state nodes; and the transition path with the minimum value of the evaluation function in the state transition graph is selected as the optimal control path.

[0011] The process parameter states corresponding to the transition probability are determined to construct a reachable state set, and a state transition graph is constructed according to the state distribution of the reachable state set and the evaluation function, including: The mean and the standard deviation of the transition probability are calculated, a state transition threshold is set according to the mean and the standard deviation, process parameter states greater than the state transition threshold are screened to construct an effective state set; identify a transition path between process parameter states based on the effective state set, mark process parameter state pairs with a transition probability greater than the state transition threshold as directly reachable states, mark process parameter state pairs with a multi-step transition probability product greater than the state transition threshold as indirectly reachable states, and form a reachable state set from the directly reachable states and the indirectly reachable states together; count the number of neighborhood states for each process parameter state in the reachable state set, calculate a density distribution of process parameter states according to the ratio of the number of neighborhood states to the total number of states in the reachable state set, and determine a distribution center of process parameter states based on the density distribution; establish a state space coordinate system with the distribution center, map the process parameter states in the reachable state set to the state space coordinate system to obtain state nodes, adjust the spacing between the state nodes according to the average transition probability between the process parameter states and their neighborhood states, and construct a process parameter state transition graph.

[0012] In a second aspect of the embodiments of the present application, an industrial production process parameter optimization and control system based on a knowledge graph is provided, which includes: A first unit is configured to collect real-time process parameter data and historical process parameter data of an industrial production device, analyze the correlation characteristics and change rules between process parameters based on the real-time process parameter data and the historical process parameter data, and construct a process parameter dynamic migration matrix. A second unit is configured to perform parameter clustering analysis on the process parameter dynamic migration matrix, identify process parameter combinations, and obtain process parameter state transition characteristics; perform dynamic evolution analysis based on the process parameter state transition characteristics, and update the process parameter dynamic migration matrix to realize real-time optimization of the correlation between process parameters. A third unit is configured to calculate an optimal parameter control path based on the process parameter dynamic migration matrix and in combination with the current state and target state of the process parameters; set a parameter optimization target according to the optimal parameter control path, and comprehensively consider the control effect of the process parameters and the change trend of the product quality to realize dynamic balance of process parameter optimization and control. A fourth unit is configured to calculate a process parameter optimization and control strategy according to the real-time process parameter data and the process parameter dynamic migration matrix, and issue the process parameter optimization and control strategy to the industrial production device to execute process parameter adjustment.

[0013] In a third aspect of the embodiments of the present application, an electronic device is provided, which includes: a processor; a memory for storing processor-executable instructions; The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0014] In a fourth aspect, the present application provides a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the method described above.

[0015] The present application has the following advantages: The present application provides a method for optimizing and regulating industrial process parameters based on a knowledge graph, which realizes precise analysis and intelligent regulation of industrial process parameters by constructing a process parameter dynamic migration matrix, avoiding the problem of inaccurate parameter regulation caused by reliance on manual experience in traditional methods.

[0016] The method dynamically updates the correlation of process parameters based on real-time and historical process parameter data, calculates the optimal regulation path, so that the process parameter optimization process can be self-adaptively adjusted with the change of the production environment, improving the stability of the industrial production process and the consistency of the product quality.

[0017] Through the construction and dynamic evolution analysis of the process parameter dynamic migration matrix, the method can effectively identify the process parameter combination characteristics and state transition rules, realize the collaborative optimization of process parameters and product quality, and significantly improve the efficiency and product quality of industrial production, and reduce energy consumption and production cost. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The figure is a flowchart of the method for optimizing and regulating industrial process parameters based on a knowledge graph according to an embodiment of the present application. Figure 2 The figure is a flowchart of the process parameter state transition graph construction according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described in detail below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] The technical scheme of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0021] Figure 1 The figure is a flowchart of the method for optimizing and regulating industrial process parameters based on a knowledge graph according to an embodiment of the present application, as shown in the figure, the method comprises: Figure 1 ​ Collecting real-time process parameter data and historical process parameter data of industrial production equipment, analyzing the correlation characteristics and change patterns between process parameters based on the real-time process parameter data and the historical process parameter data, and constructing a process parameter dynamic migration matrix; Performing parameter cluster analysis on the process parameter dynamic migration matrix to identify process parameter combinations and obtain process parameter state transition characteristics; performing dynamic evolution analysis based on the process parameter state transition characteristics and updating the process parameter dynamic migration matrix to achieve real-time optimization of process parameter association relationships; Based on the process parameter dynamic migration matrix, combined with the current state and target state of the process parameters, the optimal parameter control path is calculated; according to the optimal parameter control path, the parameter optimization target is set, and the control effect of the process parameters and the changing trend of product quality are comprehensively considered to achieve a dynamic balance between process parameter optimization and control; A process parameter optimization and control strategy is calculated based on the real-time process parameter data and the process parameter dynamic migration matrix, and the process parameter optimization and control strategy is sent to industrial production equipment to execute process parameter adjustment.

[0022] In an optional embodiment, based on the real-time process parameter data and the historical process parameter data, analyzing the correlation characteristics and change patterns between the process parameters to construct a process parameter dynamic migration matrix includes: Calculating a time-lag correlation coefficient between process parameters based on the real-time process parameter data and the historical process parameter data; The process parameters are grouped according to the values ​​of the time-lag correlation coefficients, the process parameters having the time-lag correlation coefficients greater than a preset time-lag threshold are divided into a first correlation group, and a process parameter state interval is constructed based on the process parameters in the strong correlation group; Based on the state interval of the process parameters, the state transition frequency of the process parameters at adjacent moments is counted, and the state transition probability of the process parameters is calculated according to the state transition frequency; the historical state transition probability and the real-time state transition probability of the process parameters are weightedly fused using the time-lag correlation coefficient to generate a dynamic migration matrix of the process parameters.

[0023] During industrial production, the system collects various process parameter data, such as temperature, pressure, and flow, in real time and also stores historical process parameter data. To analyze the correlation characteristics and changing patterns between process parameters and construct a dynamic process parameter migration matrix, the system first needs to calculate the time-lag correlation coefficient between the process parameters.

[0024] The calculation process of the time-lag correlation coefficient involves processing of the process parameter data. Taking a certain chemical production line as an example, assume that three process parameters of temperature, pressure, and flow rate are collected. The system takes the time series T(t) of the temperature parameter T and the time series P(t) of the pressure parameter P, and calculates the correlation after time shifting the two time series. Specifically, the system calculates the correlation coefficient of T(t) and P(t+τ) at different time lag values τ, with τ ranging from 0 to 24 hours and an interval of 1 hour. When τ = 3 hours, the correlation coefficient reaches a maximum value of 0.85, indicating that temperature changes will have a significant impact on pressure after 3 hours. Similarly, the system calculates the time-lag correlation coefficients between temperature and flow rate, and between pressure and flow rate, which reach maximum values of 0.78 and 0.65 at τ = 2 hours and τ = 1 hour, respectively.

[0025] According to the calculated time-lag correlation coefficients, the system groups the process parameters. Set the preset time-lag threshold value to 0.7, and divide the process parameters with time-lag correlation coefficients greater than 0.7 into the first correlation group. In the above example, the time-lag correlation coefficient between temperature and pressure is 0.85, and the time-lag correlation coefficient between temperature and flow rate is 0.78, both of which are greater than the threshold value of 0.7, so temperature, pressure, and flow rate are divided into the first correlation group. For other process parameters, such as stirring speed, the time-lag correlation coefficients with the above three parameters are all less than 0.7, so they are not included in the first correlation group.

[0026] Based on the process parameters in the first correlation group, the system constructs the process parameter state intervals. Taking the temperature parameter as an example, by analyzing the historical data, the system determines that the temperature varies in the range of 50-150°C, and divides it into 5 state intervals: state 1 (50-70°C), state 2 (70-90°C), state 3 (90-110°C), state 4 (110-130°C), and state 5 (130-150°C). Similarly, the system divides the pressure parameter into 4 state intervals: state 1 (0.5-1.0 MPa), state 2 (1.0-1.5 MPa), state 3 (1.5-2.0 MPa), and state 4 (2.0-2.5 MPa); and divides the flow rate parameter into 3 state intervals: state 1 (100-200 L / min), state 2 (200-300 L / min), and state 3 (300-400 L / min).

[0027] Based on the process parameter state intervals, the system counts the frequency of state transitions between adjacent moments. Taking temperature parameters as an example, the system analyzes temperature data over a 24-hour period and counts the number of transitions from state 1 to states 1, 2, 3, 4, and 5 as 80, 20, 0, 0, and 0, respectively. The number of transitions from state 2 to states 1, 2, 3, 4, and 5 is 15, 70, 15, 0, and 0, respectively. This continues in this manner, completing the statistics of transition frequencies between all states.

[0028] Based on the state transition frequency, the system calculates the state transition probability of the process parameters. For the temperature parameter, the probability of transitioning from state 1 to state 2 is 20 / (80+20+0+0+0)=0.2, and the probability of transitioning from state 1 to state 3, state 4, and state 5 is 0. The probability of transitioning from state 2 to state 1 is 15 / (15+70+15+0+0)=0.15, the probability of transitioning from state 2 to state 2 is 70 / (15+70+15+0+0)=0.7, the probability of transitioning from state 2 to state 3 is 15 / (15+70+15+0+0)=0.15, and the probability of transitioning from state 2 to state 4 and state 5 is 0. This is how the complete state transition probability matrix is ​​obtained.

[0029] The system performs the above calculations on both historical and real-time data, generating historical and real-time state transition probability matrices. Taking the temperature parameter as an example, assume the probability of transitioning from state 2 to state 3 is 0.15 in the historical data and 0.25 in the real-time data. The time-lag correlation coefficient between temperature and pressure is 0.85. The system uses this time-lag correlation coefficient to perform a weighted fusion of the historical and real-time state transition probabilities, calculating the dynamic transition probability from state 2 to state 3 as 0.15 × (1 - 0.85) + 0.25 × 0.85 = 0.2325. A similar method is used to calculate the dynamic transition probabilities between all states, ultimately generating the dynamic transition matrix for the temperature parameter.

[0030] Similarly, the system generates dynamic migration matrices for pressure and flow parameters. These matrices reflect the inherent patterns of process parameter changes over time and can be used to predict future trends, providing an important basis for production process optimization and anomaly detection.

[0031] Through the above implementation methods, the present invention achieves the technical goal of analyzing the correlation characteristics and change laws between parameters based on real-time and historical process parameter data and constructing a dynamic migration matrix, providing an effective technical means for the intelligent control of industrial production processes.

[0032] In an optional embodiment, performing parameter cluster analysis on the process parameter dynamic migration matrix, identifying process parameter combinations, and obtaining process parameter state transition characteristics includes: Constructing a state transition sequence similarity of process parameters, using the state transition sequence similarity as an initial clustering distance, and updating the initial clustering distance using a Lance-Williams parameter to obtain a clustering distance matrix of process parameters; calculating the intra-class closeness and inter-class separation of process parameter combinations based on the clustering distance matrix; Calculating an importance score of a process parameter combination according to the intra-class tightness and the inter-class separation, and screening a process parameter combination based on the importance score; performing stability verification on the process parameter combination to obtain a stability index of the process parameter combination; A main transfer path of the process parameter combination is extracted based on the stability index, a transfer rate of the main transfer path is calculated, and the transfer rate, the stability index, and the importance score are used as state transfer features of the process parameter combination.

[0033] Obtain the state transition sequence of process parameters. In a specific example, data on the state changes of 10 process parameters on a production line over a 30-day period is collected. Each process parameter has a state value on each day. For example, the temperature parameter has a state value of "high" on the first day, "medium" on the second day, "low" on the third day, and so on. For each process parameter, a state transition sequence of length 30 is obtained.

[0034] The Levenshtein distance algorithm is used to calculate the similarity between two process parameter state transition sequences. For example, the state transition sequence of process parameter A is "high-medium-low-medium-high" and the state transition sequence of process parameter B is "high-high-medium-low-medium". By calculating the minimum number of operations required to transform one sequence into the other, the Levenshtein distance between them is 2. After normalizing the Levenshtein distance, the state transition sequence similarity is 0.6. Similarity is calculated for all pairs of process parameters to form the initial cluster distance matrix.

[0035] The hierarchical clustering method is used when updating the initial clustering distance using the Lance-Williams parameter update. In each round of clustering, the two parameters or parameter clusters with the smallest distance are found to be merged, and then the distance of the new cluster to other clusters is updated. In the specific calculation, assuming that clusters P and Q have been formed, the new distance between them and cluster R can be calculated by multiplying the distance between cluster P and cluster R by the weight coefficient of cluster P, adding the distance between cluster Q and cluster R multiplied by the weight coefficient of cluster Q, and then adding the distance between cluster P and cluster Q multiplied by an interaction coefficient to obtain the new distance between cluster (P, Q) and cluster R. In actual operation, the weight coefficient can be set as the size ratio of the cluster, such as cluster P containing 3 parameters and cluster Q containing 2 parameters, and the weight coefficients are 0.6 and 0.4 respectively. Through multiple rounds of iteration and update, the complete clustering distance matrix is finally obtained.

[0036] Based on the clustering distance matrix, the intra-class compactness and inter-class separation of the process parameter combination are calculated. The intra-class compactness refers to the average distance between all pairs of parameters in a cluster. In the actual case, a cluster containing parameters A, B, and C has intra-class distances of d(A, B) = 0.3, d(A, C) = 0.4, and d(B, C) = 0.35, and the intra-class compactness of the cluster is 0.35. The inter-class separation refers to the average distance between a cluster and all other clusters. For example, the distance between cluster 1 and cluster 2 is 0.8, and the distance between cluster 1 and cluster 3 is 0.75, and the inter-class separation of cluster 1 is 0.775. The ideal clustering result should have a lower intra-class compactness and a higher inter-class separation, indicating that the similarity of parameters within the cluster is high and the difference between parameters in different clusters is large.

[0037] The importance score of the process parameter combination is calculated according to the intra-class compactness and inter-class separation. The importance score can be obtained by dividing the inter-class separation by the intra-class compactness. The higher the score, the higher the discrimination of the process parameter combination, and the more significant the impact on the production process. For example, the intra-class compactness of a certain parameter combination is 0.3, and the inter-class separation is 0.75, so the importance score is 2.5. Set the threshold to 2.0, and filter out the process parameter combinations with importance scores greater than the threshold. In this example, three process parameter combinations are selected, namely {temperature, pressure, flow rate}, {humidity, time}, and {speed, angle}.

[0038] The selected process parameter combinations are verified for stability, mainly through a cross-validation method. The 30-day data is divided into three subsets, and two subsets are used for clustering each time, and the remaining one subset is used for verification. If the clustering results obtained on different subsets are highly consistent, it is considered that the process parameter combination has high stability. The stability index can be measured by the consistency coefficient, ranging from 0 to 1. In the example, the stability index of the {temperature, pressure, flow rate} combination is 0.85, the stability index of the {humidity, time} combination is 0.78, and the stability index of the {speed, angle} combination is 0.72. Set the stability threshold to 0.75, and retain the {temperature, pressure, flow rate} and {humidity, time} combinations.

[0039] Based on the stability index, the main transition path of the process parameter combination is extracted. For the {temperature, pressure, flow rate} combination, the state changes in 30 days are analyzed, and the frequencies of various state transitions are counted. For example, the number of times from state A to state B is 8, the number of times from state B to state C is 6, and the number of times from state C to state A is 5. The path with a frequency exceeding 20% of the total transition times is determined as the main transition path, and in this example, the main transition path is the circular path A--B--C--A. The transition rate of the main transition path is calculated, that is, the number of state transitions per unit time. In this example, the transition rate of A--B is 0.27 times / day, the transition rate of B--C is 0.2 times / day, and the transition rate of C--A is 0.17 times / day. Similarly, the main transition path and transition rate of the {humidity, time} combination are calculated.

[0040] The transition rate, stability index, and importance score are used as the state transition characteristics of the process parameter combination. For the {temperature, pressure, flow rate} combination, the state transition characteristics are: the main transition path A--B--C--A, the transition rates are 0.27 times / day, 0.2 times / day, and 0.17 times / day, the stability index is 0.85, and the importance score is 2.5. For the {humidity, time} combination, the state transition characteristics are: the main transition path D--E--D, the transition rates are 0.23 times / day and 0.25 times / day, the stability index is 0.78, and the importance score is 2.2. These characteristics can be used for subsequent process parameter optimization and anomaly detection.

[0041] In an alternative embodiment, the Lance-Williams parameter is used to update the initial clustering distance to obtain the clustering distance matrix of the process parameters, including: The Lance-Williams parameter is constructed according to the number of samples in the process parameter category, and the Lance-Williams parameter is calculated by the number of samples in the two process parameter categories to be merged; updating the initial cluster distance according to the Lance-Williams parameter, performing a weighted combination of the Lance-Williams parameter and the initial cluster distance to obtain an updated distance after the process parameter categories are merged; A clustering distance matrix of process parameters is generated according to the updated distance, the clustering distance matrix records the distance relationship between process parameter categories, and the symmetry of the clustering distance matrix is ​​verified.

[0042] By constructing Lance-Williams parameters and applying them to the update of initial cluster distances, effective cluster analysis of process parameters is achieved. During implementation, the system first acquires raw process parameter data, including key parameters such as temperature, pressure, and time, which significantly impact final product quality. The system then preprocesses this raw data, including missing value handling, outlier detection and removal, and data standardization, to ensure data quality.

[0043] The system calculates the initial cluster distance for the preprocessed process parameter data. During this calculation, the Euclidean distance is used to calculate the similarity between any two process parameters, A and B. For example, when the values ​​of parameter A are [35.2, 4.5, 120] and the values ​​of parameter B are [37.8, 4.2, 115], the system calculates the Euclidean distance between these two sets of parameters, obtaining an initial distance value of 5.97. The system performs similar calculations for all pairs of process parameters to construct a complete initial distance matrix.

[0044] After obtaining the initial clustering distance, the system constructs the Lance-Williams parameter. The Lance-Williams parameter is calculated based on the number of samples in the two process parameter categories to be merged. Assuming that there are currently categories C1, C2 and C3, where C1 contains 10 samples and C2 contains 15 samples, and they need to be merged to form a new category C12, the system constructs the Lance-Williams parameter based on the ratio of the number of samples. Specifically, the α1 value is the number of C1 samples divided by the total number of C1 and C2 samples, and the calculation result is α1=10 / (10+15)=0.4; the α2 value is the number of C2 samples divided by the total number of C1 and C2 samples, and the calculation result is α2=15 / (10+15)=0.6; the β parameter varies according to the selected clustering algorithm, and is 0 in the full connection method of this embodiment; the γ parameter is 0.5 in this embodiment.

[0045] The system updates the initial cluster distance based on the Lance-Williams parameters calculated above. During the update process, the system performs a weighted combination of the Lance-Williams parameters and the initial cluster distance to calculate the distance between the merged new category and other categories. Taking the distance calculation between the merged category C12 and category C3 as an example, assuming that the initial distance between C1 and C3 is d13=8.2, and the initial distance between C2 and C3 is d23=6.5, the system calculates the updated distance between C12 and C3 according to the Lance-Williams formula d(12)3=α1×d13+α2×d23+β×|d13-d23|+γ×(d13+d23). Substituting specific values: d(12)3=0.4×8.2+0.6×6.5+0×|8.2-6.5|+0.5×(8.2+6.5)=10.35. Through similar calculations, the system updates the distance between the new category after the merger and all other categories.

[0046] The system generates a cluster distance matrix for the process parameters based on the updated distance values. This matrix records the distance relationships between all process parameter categories. For example, after a merge operation, the system-generated cluster distance matrix contains distance relationships between four categories: d12 = 5.97, d13 = 8.2, d14 = 10.5, d23 = 6.5, d24 = 9.8, and d34 = 7.3. These distance values ​​reflect the similarity between different process parameter categories, with smaller distances indicating greater similarity between the parameter categories.

[0047] To ensure the effectiveness of cluster analysis, the system performs a symmetry check on the generated cluster distance matrix. Symmetry verification involves checking whether any two elements dij and dji in the matrix are equal. The system iterates through each element in the matrix, checking whether the values ​​at symmetrical positions are consistent. For example, it checks whether d13 = d31, d24 = d42, and so on. If any inconsistency is found, the system records the error and makes adjustments. In a real-world application, a verification found that d25 = 8.7 and d52 = 8.8. After detecting this inconsistency, the system took the average value of 8.75 and corrected the corresponding elements in the matrix to ensure the symmetry of the distance matrix.

[0048] After the above steps, the system successfully constructs and verifies the clustering distance matrix of process parameters. This matrix accurately reflects the distance relationship between different process parameter categories, providing a solid foundation for subsequent clustering analysis and process parameter optimization. In this way, the system can identify similar process parameter combinations and design better process flows based on these combinations, significantly improving product quality and production efficiency. In practical applications, this method has been successfully applied to process parameter clustering analysis in various production scenarios, such as a semiconductor manufacturing process. The system identified three key parameter combinations, and through optimization of these parameter combinations, product yield improved by 8.7% and production efficiency improved by 12.3%.

[0049] In an alternative embodiment, based on the process parameter dynamic migration matrix, the optimal parameter control path is calculated by combining the current state and target state of the process parameters, which includes: An evaluation function is constructed based on the transition probabilities in the process parameter dynamic migration matrix, which reflects the transition difficulty between different process parameter states. Transition probabilities greater than a preset migration threshold are identified from the process parameter dynamic migration matrix to determine the process parameter states corresponding to the transition probabilities and construct a reachable state set. A state transition graph is constructed based on the state distribution of the reachable state set and the evaluation function. The state node intervals in the state transition graph are calibrated, the transition difficulty between adjacent state nodes is calculated based on the evaluation function, and the transition difficulty is calibrated between the corresponding state nodes. The transition path with the smallest value of the evaluation function in the state transition graph is selected as the optimal control path.

[0050] A process parameter dynamic migration matrix is obtained, which records the transition probabilities between different process parameter states. These transition probabilities reflect the difficulty of transitioning from one process parameter state to another. Specifically, each element P(i,j) in the matrix represents the probability of transitioning from state i to state j. For example, if the success rate of adjusting the temperature parameter from 350°C to 380°C in a production process is 85%, the value at the corresponding position in the process parameter dynamic migration matrix is 0.85.

[0051] When constructing the evaluation function based on the process parameter dynamic migration matrix, a negative logarithm conversion method can be used to convert the transition probability into a transition difficulty indicator. For any two states i and j, the transition difficulty D(i,j) can be represented as the negative logarithm value of the transition probability P(i,j). When the transition probability is higher, the corresponding transition difficulty is lower; conversely, when the transition probability is lower, the corresponding transition difficulty is higher. For example, if the transition probability from state A to state B is 0.9, the transition difficulty is 0.105; while the transition probability from state A to state C is 0.3, the transition difficulty is 1.204, which is significantly higher than the former.

[0052] When identifying the transition probability greater than the preset transition threshold from the process parameter dynamic transition matrix, the threshold value can be set as 0.6. When the transition probability between two states is greater than 0.6, it is considered that the two states can be directly transitioned. By traversing the entire transition matrix, all state pairs that meet the conditions can be found. Assuming that there are 10 states in a process parameter space, after threshold screening, 23 pairs of directly transitionable state pairs are determined, which constitute the reachable state set.

[0053] When constructing the state transition graph according to the reachable state set, each process parameter state is represented as a node in the graph, and a direct transitionable state pair is connected by an edge. For example, if state 1 and state 2 can be directly transitioned, then node 1 and node 2 are connected in the transition graph. In this way, a state transition graph containing all 10 state nodes and 23 connection edges can be obtained.

[0054] When calibrating the state node interval in the state transition graph, the physical distance and transition difficulty between states need to be considered. The physical distance can be the difference in process parameter values, for example, the physical distance of the temperature parameter from 350°C to 380°C is 30°C. The transition difficulty comes from the evaluation function constructed earlier. For adjacent state nodes i and j, the calibration value between them can be considered by combining the physical distance and the transition difficulty, that is, the transition difficulty D(i,j) is marked on the edge connecting nodes i and j.

[0055] In order to actually calculate the optimal control path, it is assumed that the current process parameter state is state 2 and the target state is state 8. By applying the Dijkstra algorithm, the transition difficulty cumulative value of the path from state 2 to state 8 can be found in the constructed state transition graph. During the calculation process, the node with the smallest transition difficulty cumulative value among the currently unvisited nodes is selected for expansion each time until the target node is reached.

[0056] In a specific case, a semiconductor manufacturing process involves three key parameters: temperature, pressure, and flow rate. The parameter values of the current state are: temperature 380°C, pressure 2.5 MPa, flow rate 15 m / s; the parameter values of the target state are: temperature 420°C, pressure 3.2 MPa, flow rate 20 m / s. Through discretization processing, the continuous parameter space is divided into a finite number of state points, resulting in 64 different state combinations. The constructed process parameter dynamic transition matrix is 64x64.

[0057] The transition probability threshold is set to 0.65, and 142 pairs of directly transitionable state pairs are obtained through screening. Based on these state pairs, a state transition graph is constructed, and an evaluation function is used to calculate the transition difficulty between each pair of adjacent states. The current state corresponds to state number 17, and the target state corresponds to state number 53.

[0058] After applying the path search algorithm, the optimal control path obtained is: 17--24--31--38--45--53. This means that the process parameters should be adjusted in the following order: first, increase the temperature from 380℃ to 390℃ while keeping other parameters unchanged; then increase the pressure from 2.5MPa to 2.7MPa; next, increase the flow rate from 15m / s to 17m / s; then further increase the temperature to 405℃; finally, adjust the temperature to 420℃, the pressure to 3.2MPa, and the flow rate to 20m / s simultaneously. The total transition difficulty value of this path is 1.87, which is significantly lower than the transition difficulty values of other paths.

[0059] The optimal control path determined by this method not only considers the physical feasibility of parameter adjustment, but also maximizes the reduction of risk and difficulty in the parameter adjustment process, thereby improving the stability and reliability of production process parameter adjustment. Practical application shows that compared with the traditional direct linear adjustment method, this method can reduce the parameter adjustment failure rate from 12% to 3.5%, significantly improving production efficiency.

[0060] In an alternative embodiment, the process parameter state corresponding to the transition probability is determined to construct a reachable state set, and the state transition graph is constructed according to the state distribution of the reachable state set and the evaluation function, which includes: The mean and standard deviation of the transition probability are calculated, the state transition threshold is set according to the mean and standard deviation, and the process parameter states greater than the state transition threshold are screened to construct an effective state set; Based on the effective state set, the transition path between process parameter states is identified, the process parameter state pair with a transition probability greater than the state transition threshold is marked as a directly reachable state, and the process parameter state pair with a multi-step transition probability product greater than the state transition threshold is marked as an indirectly reachable state, and the directly reachable state and the indirectly reachable state together form a reachable state set; The number of neighborhood states of each process parameter state in the reachable state set is counted, the density distribution of the process parameter state is calculated according to the ratio of the number of neighborhood states to the total number of states in the reachable state set, and the distribution center of the process parameter state is determined based on the density distribution; A state space coordinate system is established based on the distribution center, the process parameter states in the reachable state set are mapped to the state space coordinate system to obtain state nodes, the spacing between the state nodes is adjusted according to the mean of the transition probability between the process parameter state and its neighborhood state, and a process parameter state transition graph is constructed.

[0061] As shown in Figure 2 The method includes: After obtaining the state transition probabilities of the process parameters, the system calculates the mean and standard deviation of these transition probabilities. For example, assume that the system collects 100 sets of process parameter state transition data and calculates that the mean of the transition probabilities is 0.65 and the standard deviation is 0.15. The system sets the state transition threshold value as the mean minus the standard deviation, i.e. 0.65-0.15=0.5, as the threshold value for screening the valid states. In the screening process, the system retains the process parameter state pairs with a transition probability greater than 0.5 to construct the valid state set. For example, the transition probability of process parameter A from state 1 to state 2 is 0.72, which is greater than the threshold value 0.5, so the state pair is included in the valid state set.

[0062] When identifying the transition path between the process parameter states based on the valid state set, the system marks the process parameter state pairs with a transition probability greater than the threshold value as directly reachable states. For example, the transition probability of process parameter B from state 3 to state 5 is 0.68, which is greater than the threshold value 0.5, so the system marks it as a directly reachable state. For state pairs without a transition probability greater than the threshold value, the system calculates the probability product of multi-step transitions. For example, there is no directly reachable state pair for process parameter C from state 2 to state 4 with a transition probability greater than the threshold value, but the transition probability from state 2 to state 3 is 0.76 and the transition probability from state 3 to state 4 is 0.82, and their product is 0.76x0.82=0.6232, which is greater than the threshold value 0.5, so the system marks state 2 to state 4 as an indirectly reachable state. The directly reachable states and the indirectly reachable states together constitute the reachable state set.

[0063] For each process parameter state in the reachable state set, the system counts the number of its neighborhood states. The neighborhood state refers to other states that have a direct or indirect reachable relationship with the current state. For example, state 7 of process parameter D has 8 neighborhood states in the reachable state set, and the total number of states in the reachable state set is 25, so the density distribution of state 7 is 8 / 25=0.32. The system calculates the density distribution of all states and finds the state with the highest density distribution as the distribution center. Assume that the density distribution of state 10 of process parameter D is 0.56, which is the highest among all states, so the system determines it as the distribution center of the process parameter states.

[0064] When establishing the state space coordinate system with the distribution center, the system sets the distribution center (state 10) as the coordinate origin (0, 0), and maps other state nodes to the coordinate system according to their relationship with the distribution center. For example, state 8 directly connected to state 10 is mapped to the coordinate (1, 0). Subsequently, the system adjusts the distance between the state nodes according to the mean of the transition probabilities between the process parameter states and their neighborhood states. The higher the mean of the transition probabilities, the easier the transition between the states occurs, and the smaller the distance between the corresponding state nodes. The lower the mean of the transition probabilities, the larger the distance between the state nodes.

[0065] The specific adjustment method is that the system calculates the average transition probability between each pair of adjacent state nodes, and converts it into a distance factor. For example, the average transition probability between state 3 and state 5 of process parameter E is 0.85, and the system can set the distance between them to (1-0.85)=0.15 times of the basic unit distance, that is, the distance is closer; while the average transition probability between state 5 and state 8 is 0.52, the distance can be set to (1-0.52)=0.48 times of the basic unit distance, which is farther than the former. In this way, the system arranges all the state nodes in the state space coordinate system according to their transition relationship, forming a process parameter state transition diagram.

[0066] In practical application, assuming that a manufacturing process has 5 key process parameters, each parameter has 3 states, and a total of 243 process parameter combination states are formed. After transition probability calculation and threshold screening, the system obtains 87 effective states. Through analysis of direct and indirect reachable relationship, the system finally constructs a state transition diagram containing 72 nodes and 118 connections, which clearly shows the transition relationship between process parameter states.

[0067] The states close to the center node in the figure represent more common or stable process states, which can be used as reference points for process optimization; while the states far from the center and with few connections represent abnormal or non-optimal process conditions, and the system can prompt the operator to avoid entering these states. Through this visual state transition diagram, process engineers can intuitively understand the transition rules between parameter states, and accordingly adjust the process flow to improve production efficiency and product quality.

[0068] The state transition diagram construction method combines probability statistics and spatial layout technology to visualize the complex process parameter state transition relationship, providing a powerful tool for process optimization. Practice has proved that the state transition diagram constructed by this method can accurately reflect the process parameter state distribution characteristics and transition rules, and has significant value for guiding actual production.

[0069] In a second aspect of the embodiment of the present application, an industrial production process parameter optimization and control system based on a knowledge graph is provided, comprising: A first unit is configured to collect real-time process parameter data and historical process parameter data of an industrial production device, analyze the correlation characteristics and change rules between process parameters based on the real-time process parameter data and the historical process parameter data, and construct a process parameter dynamic migration matrix. A second unit is configured to perform parameter clustering analysis on the process parameter dynamic migration matrix, identify process parameter combinations, and obtain process parameter state transition characteristics; perform dynamic evolution analysis based on the process parameter state transition characteristics, and update the process parameter dynamic migration matrix to realize real-time optimization of the correlation relationship between process parameters. The third unit is configured to calculate an optimal parameter control path based on the process parameter dynamic migration matrix, the current state and the target state of the process parameter, set a parameter optimization target according to the optimal parameter control path, and comprehensively consider the control effect of the process parameter and the change trend of the product quality to realize dynamic balance of process parameter optimization and control. The fourth unit is configured to calculate a process parameter optimization control strategy according to the real-time process parameter data and the process parameter dynamic migration matrix, and send the process parameter optimization control strategy to an industrial production device to execute process parameter adjustment.

[0070] In a third aspect, the present application provides an electronic device, comprising: a processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described above.

[0071] In a fourth aspect, the present application provides a computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions are executed by a processor to implement the method described above.

[0072] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein, which are used to execute various aspects of the present application.

[0073] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. The industrial production process parameter optimization and control method based on knowledge graph is characterized by: include: Collecting real-time process parameter data and historical process parameter data of industrial production equipment, analyzing the correlation characteristics and change patterns between process parameters based on the real-time process parameter data and the historical process parameter data, and constructing a process parameter dynamic migration matrix; Performing parameter cluster analysis on the process parameter dynamic migration matrix to identify process parameter combinations and obtain process parameter state transition characteristics; Performing dynamic evolution analysis based on the state transition characteristics of the process parameters and updating the process parameter dynamic migration matrix to achieve real-time optimization of the process parameter association relationship; Based on the process parameter dynamic migration matrix, combined with the current state and target state of the process parameters, the optimal parameter control path is calculated; according to the optimal parameter control path, the parameter optimization target is set, and the control effect of the process parameters and the changing trend of product quality are comprehensively considered to achieve a dynamic balance between process parameter optimization and control; A process parameter optimization and control strategy is calculated based on the real-time process parameter data and the process parameter dynamic migration matrix, and the process parameter optimization and control strategy is sent to industrial production equipment to execute process parameter adjustment.

2. The method according to claim 1, characterized in that Analyzing the correlation characteristics and change patterns between the process parameters based on the real-time process parameter data and the historical process parameter data, and constructing a process parameter dynamic migration matrix includes: Calculating a time-lag correlation coefficient between process parameters based on the real-time process parameter data and the historical process parameter data; The process parameters are grouped according to the values ​​of the time-lag correlation coefficients, the process parameters having the time-lag correlation coefficients greater than a preset time-lag threshold are divided into a first correlation group, and a process parameter state interval is constructed based on the process parameters in the strong correlation group; Based on the state interval of the process parameters, the state transition frequency of the process parameters at adjacent moments is counted, and the state transition probability of the process parameters is calculated according to the state transition frequency; the historical state transition probability and the real-time state transition probability of the process parameters are weightedly fused using the time-lag correlation coefficient to generate a dynamic migration matrix of the process parameters.

3. The method according to claim 1, characterized in that Performing parameter cluster analysis on the process parameter dynamic migration matrix, identifying process parameter combinations, and obtaining process parameter state transition characteristics includes: Constructing a state transition sequence similarity of process parameters, using the state transition sequence similarity as an initial clustering distance, and updating the initial clustering distance using a Lance-Williams parameter to obtain a clustering distance matrix of process parameters; calculating the intra-class closeness and inter-class separation of process parameter combinations based on the clustering distance matrix; Calculating an importance score of a process parameter combination according to the intra-class tightness and the inter-class separation, and screening a process parameter combination based on the importance score; performing stability verification on the process parameter combination to obtain a stability index of the process parameter combination; A main transfer path of the process parameter combination is extracted based on the stability index, a transfer rate of the main transfer path is calculated, and the transfer rate, the stability index, and the importance score are used as state transfer features of the process parameter combination.

4. The method according to claim 3, characterized in that The initial cluster distance is updated using the Lance-Williams parameter to obtain a cluster distance matrix of process parameters including: constructing a Lance-Williams parameter based on the number of samples in the process parameter category, wherein the Lance-Williams parameter is calculated using the number of samples in the two process parameter categories to be merged; updating the initial cluster distance according to the Lance-Williams parameter, performing a weighted combination of the Lance-Williams parameter and the initial cluster distance to obtain an updated distance after the process parameter categories are merged; A clustering distance matrix of process parameters is generated according to the updated distance, the clustering distance matrix records the distance relationship between process parameter categories, and the symmetry of the clustering distance matrix is ​​verified.

5. The method according to claim 1, wherein Based on the process parameter dynamic migration matrix, combined with the current state and target state of the process parameters, the optimal parameter control path is calculated, including: An evaluation function is constructed based on the transition probabilities in the process parameter dynamic migration matrix, where the evaluation function reflects the difficulty of transitions between different process parameter states; a transition probability greater than a preset migration threshold is identified from the process parameter dynamic migration matrix, the process parameter states corresponding to the transition probabilities are determined to construct a reachable state set, and a state transition diagram is constructed based on the state distribution of the reachable state set and the evaluation function; The state node intervals in the state transition diagram are calibrated, the transfer difficulty between adjacent state nodes is calculated based on the evaluation function, and the transfer difficulty is calibrated between corresponding state nodes; and the transfer path with the smallest value of the evaluation function is selected in the state transition diagram as the optimal control path.

6. The method according to claim 5, characterized in that Determining the process parameter state corresponding to the transition probability to construct a reachable state set, and constructing a state transition diagram according to the state distribution of the reachable state set and the evaluation function includes: Calculating the mean and standard deviation of the transition probability, setting a state transition threshold according to the mean and the standard deviation, and screening process parameter states greater than the state transition threshold to construct a valid state set; Identify transition paths between process parameter states based on the valid state set, mark process parameter state pairs with transition probabilities greater than the state transition threshold as directly reachable states, mark process parameter state pairs with multi-step transition probabilities greater than the state transition threshold as indirectly reachable states, and form a reachable state set based on the directly reachable states and the indirectly reachable states; For each process parameter state in the reachable state set, counting the number of neighboring states, calculating the density distribution of the process parameter state according to a ratio of the number of neighboring states to the total number of states in the reachable state set, and determining the distribution center of the process parameter state based on the density distribution; A state space coordinate system is established with the distribution center, and the process parameter states in the reachable state set are mapped to the state space coordinate system to obtain state nodes. The spacing between the state nodes is adjusted according to the mean value of the transition probability between the process parameter state and its neighboring states to construct a process parameter state transition diagram.

7. An industrial production process parameter optimization and control system based on knowledge graph, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to collect real-time process parameter data and historical process parameter data of industrial production equipment, analyze the correlation characteristics and change patterns between process parameters based on the real-time process parameter data and the historical process parameter data, and construct a process parameter dynamic migration matrix; The second unit is used to perform parameter cluster analysis on the process parameter dynamic migration matrix, identify process parameter combinations, and obtain process parameter state transition characteristics; Performing dynamic evolution analysis based on the state transition characteristics of the process parameters and updating the process parameter dynamic migration matrix to achieve real-time optimization of the process parameter association relationship; The third unit is configured to calculate an optimal parameter control path based on the process parameter dynamic migration matrix and in combination with the current state and target state of the process parameters; set a parameter optimization target according to the optimal parameter control path, comprehensively consider the control effect of the process parameters and the changing trend of product quality, and achieve a dynamic balance between process parameter optimization and control; The fourth unit is used to calculate the process parameter optimization and control strategy based on the real-time process parameter data and the process parameter dynamic migration matrix, and send the process parameter optimization and control strategy to the industrial production equipment to perform process parameter adjustment.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images