A production monitoring method and system of a non-toxic household disinfectant and a storage medium

Abstract

The application relates to the technical field of production monitoring, and discloses a production monitoring method and system for non-toxic household disinfectant and a storage medium. The method comprises the following steps: acquiring a parameter configuration group of household disinfectant, defining key indicators for evaluating the quality of the household disinfectant, measuring the key indicators of each production batch to obtain historical measurement data, reducing the dimension of the historical configuration data to obtain key parameters; establishing a performance prediction model and an anomaly analysis model, inputting real-time configuration data of the key parameters in the current batch into the performance prediction model to obtain the batch type of the current batch, if the current batch is an abnormal batch, obtaining a dangerous parameter from the key parameters based on the anomaly analysis model, and obtaining an abnormal reason based on the dangerous parameter; constructing a deep reinforcement learning model, iteratively optimizing the dangerous parameter, and inputting the optimized dangerous parameter into the performance prediction model until it is determined that the current batch is a normal batch. The application improves the production efficiency of the household disinfectant.

Description

Technical Field

[0001] This application relates to the field of production monitoring technology, and in particular to a production monitoring method, system and storage medium for a non-toxic household disinfectant. Background Technology

[0002] With increasing emphasis on household hygiene and health, the market demand for non-toxic, non-irritating, and environmentally friendly household disinfectants is growing. However, the production process of disinfectants is complex, involving the precise proportioning of various chemical components and the control of reaction conditions. Any slight deviation can lead to product quality problems or even the generation of toxic substances. Therefore, how to achieve precise monitoring and control of the production process of non-toxic household disinfectants to ensure stable product quality and non-toxicity has become an urgent problem to be solved by the industry.

[0003] Currently, there are few methods for monitoring the production of household disinfectants. However, methods for monitoring the production of other items often utilize machine learning algorithms. For example, Chinese patent application CN118963303A discloses an intelligent monitoring method and system for milk powder production lines based on artificial intelligence. This method acquires the status data of the target production line equipment and identifies target anomaly nodes reflecting the target anomaly monitoring items, generating a corresponding first state path vector. Then, based on the current and next equipment's production line equipment status data, it acquires the status data of the anomaly monitoring items, generating a corresponding second state path vector. This second state path vector is then fused with the first state path vector to output the target state path vector. The anomaly index of the anomaly monitoring item is determined, and the linkage feature parameters of this anomaly index in the next equipment's status data are obtained. This allows for the identification of the anomaly monitoring item nodes reflecting the target anomaly monitoring item's anomaly state partition in the next equipment's status data, thereby predicting potential anomalies.

[0004] For example, Chinese patent application CN119004214A discloses a method, system, and medium for correcting printing parameters of a solder paste printer. The method includes acquiring historical operating parameters and historical environmental parameters of the solder paste printer within a preset time period before the occurrence of frequent drift printing parameters; constructing a printing parameter drift prediction model based on the historical operating parameters and historical environmental parameters of the hazardous parameters; acquiring real-time operating parameters of the solder paste printer and real-time environmental parameters of the printing environment within the preset working time period; importing the real-time operating parameters and real-time environmental parameters of the hazardous parameters into the printing parameter drift prediction model for prediction; analyzing the prediction results to obtain the drift warning printing parameters of the solder paste printer; and generating a printing parameter correction scheme by combining the drift warning printing parameters of the solder paste printer.

[0005] However, the aforementioned existing technologies are inefficient in detecting production process anomalies and in optimizing parameters, and cannot effectively monitor the production process. Summary of the Invention

[0006] This application provides a production monitoring method, system, and storage medium for non-toxic household disinfectant, which can improve the production efficiency of non-toxic household disinfectant.

[0007] In a first aspect, this application provides a method for monitoring the production of a non-toxic household disinfectant, the method comprising: Step S1: Obtain historical configuration data of household disinfectant during the production process. The historical configuration data includes various configuration parameters. Define the combination of various configuration parameters for each production batch as a parameter configuration group and define key indicators for evaluating the quality of the household disinfectant. Step S2: Measure the key indicators of each production batch of household disinfectant to obtain historical measurement data. The historical measurement data and the corresponding parameter configuration group together constitute the first dataset. Normalize and denoise the first dataset. Based on the first dataset, reduce the dimensionality of the historical configuration data to obtain the key parameters. Step S3: Establish a performance prediction model and anomaly analysis model, obtain real-time configuration data of key parameters of the current batch of household disinfectant during the production process and input them into the performance prediction model, the performance prediction model outputs the batch type of the current batch, if the current batch is an abnormal batch, then obtain the dangerous parameters from the key parameters based on the anomaly analysis model, the key parameters other than the dangerous parameters are defined as ordinary parameters, and obtain the cause of the anomaly based on the dangerous parameters. Step S4: Construct a deep reinforcement learning model, iteratively optimize the dangerous parameters, and input the optimized dangerous parameters and the normal parameters into the performance prediction model until the current batch is determined to be a normal batch.

[0008] In conjunction with the first aspect, in the first implementation of the first aspect of this application, the historical configuration data is dimensionality reduced based on the first dataset to obtain key parameters, including: Set a standard value for each configuration parameter, calculate the matching degree between the historical configuration data of each configuration parameter and the standard value, normalize the historical configuration data of the configuration parameters and the matching degree for each production batch, and estimate the quality of the i-th production batch of household disinfectant based on the first formula to obtain the corresponding estimated value. The first formula is: , The total number of configuration parameters. Let k be the proportional constant of the k-th configuration parameter in the i-th production batch. This represents the degree of matching between the historical configuration data of the k-th configuration parameter in the i-th production batch and the standard value. This refers to the k-th configuration parameter in the i-th production batch. For each configuration parameter For the estimated value The degree of contribution; If the estimated value is greater than the first threshold, then the configuration parameters whose contribution is greater than the second threshold are selected from all the configuration parameters corresponding to the estimated value as key parameters.

[0009] In conjunction with the first aspect, in the second implementation of the first aspect of this application, calculating the degree of matching between the historical configuration data of each configuration parameter and the standard value includes: The matching degree between the historical configuration data of the k-th configuration parameter in the i-th production batch and the standard value is calculated based on the second formula. The second formula is: ,in, This represents the actual value of the k-th configuration parameter in the i-th production batch. This is the standard value for the kth configuration parameter.

[0010] In conjunction with the first aspect, in the third implementation of the first aspect of this application, the performance prediction model outputs the batch type of each production batch, including: the performance prediction model is a regression model, the regression model generates predicted values ​​for each key indicator, sets a standard range for each key indicator, determines whether each predicted value is within the standard range, if so, defines the key indicator corresponding to the predicted value as a reliable prediction indicator, if the total number of key indicators that are reliable prediction indicators in the production batch is greater than or equal to a third threshold, then the production batch is defined as a normal batch, otherwise it is defined as an abnormal batch.

[0011] In conjunction with the first aspect, in the fourth implementation of the first aspect of this application, obtaining a dangerous parameter from the key parameters based on the cause of the anomaly, and obtaining the cause of the anomaly based on the dangerous parameter, includes: Collect simulated configuration data of key parameters corresponding to all abnormal causes, extract multiple feature quantities from the simulated configuration data, and define the abnormal causes and the feature quantities as a second dataset; A first random forest model is created, and all the second datasets are input into the first random forest model for training. The first random forest model calculates the change in Gini coefficient before and after splitting for each feature in each decision tree. Taking the root node of each decision tree as the starting node, it is determined whether there is a path in the decision tree with the feature as the splitting node. If so, the decision tree is defined as the first decision tree. The sum of the changes in Gini coefficients corresponding to all first decision trees and the total number of first decision trees are calculated. The ratio of the sum of the changes in Gini coefficients to the total number of trees is defined as the risk index of the feature. The risk indices of all features corresponding to the abnormal cause are arranged in order, and the key parameters corresponding to the features ranked above the fourth threshold are defined as risk parameters. A second random forest model is created. The simulated configuration data corresponding to the hazard parameter and the corresponding anomaly cause are input into the second random forest model. The random forest model outputs the corresponding anomaly cause. The real-time configuration data is input into the second random forest model. The second random forest model outputs the anomaly cause for the current batch.

[0012] In conjunction with the first aspect, in the fifth implementation of the first aspect of this application, the hazardous parameter is iteratively optimized, including: Multiple hazard parameters are combined to generate a state vector. The adjusted value is defined as the adjustment amount of the hazard parameters. The state vector is used as the state input of the deep reinforcement learning model, and the adjustment amount of the hazard parameters is used as the action output of the deep reinforcement learning model. The reward function of the deep learning model is set as follows: ,in The adjustment amount of the hazard parameter taken under state vector s The reward value is denoted by m, where m is the total number of the key indicators. Let be the weight parameter for the i-th key indicator. The predicted values ​​of key indicators obtained by the performance prediction model are input to the state vector s, with the adjustment amount of the danger parameter being q. Let i be the standard value of the i-th key indicator. To adjust the coefficient, the adjustment amount of the danger parameter in each abnormal batch is dynamically adjusted within a preset range until the reward function reaches its maximum value.

[0013] In conjunction with the first aspect, in the sixth implementation of the first aspect of this application, the adjustment amount of the hazardous parameter in each abnormal batch is dynamically adjusted within a preset range, including: Obtain the minimum standard for each key indicator, and calculate the adjustment amount of the hazard parameter at time t based on the third formula. The third formula is: ,in, This represents the preset optimal adjustment amount for the danger parameter at time t. The weighting coefficients for the optimal adjustment amount. The weighting coefficients for the optimal adjustment amount are obtained based on the fourth formula, which is: (The original text contains some formatting errors and inconsistencies. A more accurate translation would require the full context.) ,in, As the initial weights, The value of the optimal adjustment obtained based on the reinforcement learning algorithm. To obtain the value of previously adjusted amounts based on reinforcement learning algorithms, The minimum standard for each key indicator, It is a preset positive number.

[0014] Secondly, this application provides a production monitoring system for a non-toxic household disinfectant, the system comprising: The acquisition module is used to acquire historical configuration data of household disinfectant during the production process. The historical configuration data includes a variety of configuration parameters. The combination of various configuration parameters for each production batch is defined as a parameter configuration group, and key indicators for evaluating the quality of the household disinfectant are defined. The processing module is used to measure the key indicators of each production batch of household disinfectant to obtain historical measurement data. The historical measurement data and the corresponding parameter configuration group together constitute a first dataset. The first dataset is normalized and denoised. Based on the first dataset, the historical configuration data is dimensionality reduced to obtain key parameters. The performance prediction and anomaly analysis module is used to establish a performance prediction model and anomaly analysis model. It obtains real-time configuration data of key parameters of the current batch of household disinfectant during the production process and inputs them into the performance prediction model. The performance prediction model outputs the batch type of the current batch. If the current batch is an abnormal batch, it obtains dangerous parameters from the key parameters based on the anomaly analysis model. Key parameters other than dangerous parameters are defined as ordinary parameters. The cause of the anomaly is obtained based on the dangerous parameters. An optimization module is used to construct a deep reinforcement learning model, iteratively optimize the dangerous parameters, and input the optimized dangerous parameters and the normal parameters into the performance prediction model until the current batch is determined to be a normal batch.

[0015] A third aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the aforementioned method for monitoring the production of a non-toxic household disinfectant.

[0016] The technical solution provided in this application collects historical configuration data, including various configuration parameters, and defines parameter configuration groups to trace the specific conditions of each production batch, facilitating analysis and optimization. It clarifies key indicators for evaluating the quality of household disinfectants, providing targets for subsequent quality control and model building. The first dataset, composed of historical measurement data and parameter configuration groups, provides complete information for model training and analysis. Normalization ensures that data of different dimensions and ranges can be compared under the same standard, improving model accuracy and generalization ability, reducing data dimensionality, extracting key parameters from configuration parameters, simplifying the model, and improving computational efficiency and prediction accuracy. The performance prediction model can output the type of the current batch in real time, quickly providing feedback on production status. The anomaly analysis model can identify abnormal batches and obtain dangerous parameters from key parameters, helping to quickly locate problems and obtain the causes of anomalies based on dangerous parameters, providing a scientific basis for problem solving. A deep reinforcement learning model automatically iterates and optimizes dangerous parameters, reducing manual intervention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of an embodiment of a production monitoring method for a non-toxic household disinfectant according to the present application. Figure 2 This is a schematic diagram of one embodiment of a production monitoring system for a non-toxic household disinfectant in this application. Detailed Implementation

[0019] This application provides a method, system, and storage medium for monitoring the production of a non-toxic household disinfectant. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the production monitoring method for a non-toxic household disinfectant in this application includes: Step S1: Obtain historical configuration data of household disinfectant during the production process. The historical configuration data includes various configuration parameters. Define the combination of various configuration parameters for each production batch as a parameter configuration group and define the key indicators for evaluating the quality of household disinfectant.

[0021] Specifically, historical configuration data of multiple production batches of household disinfectant during the production process is extracted from the production database. The historical configuration data includes various configuration parameters, such as raw material ratios, production process parameters (such as temperature, pressure, stirring speed, reaction time, etc.), environmental parameters (such as humidity, cleanliness, etc.), and equipment status parameters (such as running, shutdown, maintenance, etc.). The collected historical configuration data should cover multiple batches under different time periods and production conditions to ensure the representativeness and diversity of the data. Parameter configuration groups include, for example, raw material ratios, reaction temperatures, and reaction times.

[0022] Key indicators for evaluating the quality of non-toxic household disinfectants include, but are not limited to: sterilization rate, pH value, viscosity, and irritation. Each key indicator is quantified. For example, sterilization rate can be represented by the degree to which the number of surviving microorganisms decreases after a certain period of time when a known number of specific microorganisms are mixed with the disinfectant.

[0023] Step S2: Measure the key indicators of each production batch of household disinfectant to obtain historical measurement data. The historical measurement data and the corresponding parameter configuration group together constitute the first dataset. Normalize and denoise the first dataset. Based on the first dataset, reduce the dimensionality of the historical configuration data to obtain the key parameters.

[0024] Specifically, at the end of each production batch, key indicators such as the sterilization rate, pH value, viscosity, and irritation of the household disinfectant are precisely measured. The historical measurement data obtained are integrated with the parameter configuration group of that batch (such as raw material ratio, temperature, time, etc.) to form the first dataset. All parameters in the first dataset are normalized, and random noise in the first dataset is reduced by identifying and processing outliers and using smoothing techniques or filters.

[0025] The process of dimensionality reduction of historical configuration data based on the first dataset to obtain key parameters will be explained later. Key parameters are configuration parameters that have a certain impact on the key indicators of household disinfectant. By identifying key parameters, we can better understand the configuration parameters with risks and deviations in the production process, take measures in advance to reduce risks, and then use the historical configuration data of key parameters to build a performance prediction model to predict the quality performance of future production batches, thereby improving work efficiency.

[0026] Step S3: Establish a performance prediction model and anomaly analysis model. Obtain real-time configuration data of key parameters of the current batch of household disinfectant during the production process and input them into the performance prediction model. The performance prediction model outputs the batch type of the current batch. If the current batch is an abnormal batch, then obtain the dangerous parameters from the key parameters based on the anomaly analysis model. Key parameters other than dangerous parameters are defined as ordinary parameters. Obtain the cause of the anomaly based on the dangerous parameters.

[0027] Specifically, both the performance prediction model and the anomaly analysis model are, for example, random forest models. Historical configuration data of key parameters are used as input data for the performance prediction model. The performance prediction model calculates the predicted value of each key indicator and outputs the production batch type based on the predicted value. Real-time configuration data of key parameters for the current batch are input into the performance prediction model, and the model outputs the prediction result for the current batch. If the current batch is an abnormal batch, the anomaly analysis model is trained using historical configuration data marked as abnormal. This model focuses on identifying dangerous parameters from the key parameters. Based on the identified dangerous parameters and combined with domain knowledge of the production process, the specific causes of the anomalies are analyzed and determined. Through these performance predictions and anomaly analyses, the production process of household disinfectant can be effectively monitored and optimized to ensure product quality, while timely detection and handling of anomalies in the production process.

[0028] Step S4: Construct a deep reinforcement learning model, iteratively optimize the hazardous parameters, and input the optimized hazardous parameters and normal parameters into the performance prediction model until the current batch is determined to be a normal batch.

[0029] Specifically, to construct a deep reinforcement learning model such as a deep Q-network, first define an environment that simulates the production process of household disinfectant. The state space includes key parameters such as raw material ratio, temperature, and time, while the action space consists of all possible adjustable hazardous parameters, which are considered to be variables that have a significant impact on the production process. In the simulated production environment, the model is trained by interacting with the environment.

[0030] The model learns by trying different actions and receiving rewards. It gradually learns which actions (adjustments to the hazard parameters) can maximize the reward value under different states, thus optimizing the production process. The hazard parameters and other key parameters optimized by the deep reinforcement learning model are input into the performance prediction model. The performance prediction model quickly evaluates the impact of the optimized parameters on product quality, obtains the batch type of the current batch, and continues until the current batch is identified as a normal batch.

[0031] The technical solution provided in this application collects historical configuration data, including various configuration parameters, and defines parameter configuration groups to trace the specific conditions of each production batch, facilitating analysis and optimization. It clarifies key indicators for evaluating the quality of household disinfectants, providing targets for subsequent quality control and model building. The first dataset, composed of historical measurement data and parameter configuration groups, provides complete information for model training and analysis. Normalization ensures that data of different dimensions and ranges can be compared under the same standard, improving model accuracy and generalization ability, reducing data dimensionality, extracting key parameters from configuration parameters, simplifying the model, and improving computational efficiency and prediction accuracy. The performance prediction model can output the type of the current batch in real time, quickly providing feedback on production status. The anomaly analysis model can identify abnormal batches and obtain dangerous parameters from key parameters, helping to quickly locate problems and obtain the causes of anomalies based on dangerous parameters, providing a scientific basis for problem solving. A deep reinforcement learning model automatically iterates and optimizes dangerous parameters, reducing manual intervention.

[0032] In one specific embodiment, the dimensionality reduction of historical configuration data based on the first dataset to obtain key parameters includes the following steps: Set a standard value for each configuration parameter, calculate the matching degree between the historical configuration data and the standard value for each configuration parameter, normalize the historical configuration data and matching degree for each production batch, and estimate the quality of the i-th production batch of household disinfectant based on the first formula to obtain the corresponding estimated value. The first formula is: , The total number of configuration parameters. Let k be the proportional constant of the k-th configuration parameter in the i-th production batch. This represents the degree of matching between the historical configuration data and the standard value of the k-th configuration parameter in the i-th production batch. This is the k-th configuration parameter in the i-th production batch. For each configuration parameter For the estimated value The extent of their contribution.

[0033] If the estimated value is greater than the first threshold, then the configuration parameters whose contribution is greater than the second threshold are selected from all the configuration parameters corresponding to the estimated value as key parameters.

[0034] Specifically, assume that the standard values ​​for the four configuration parameters in production batch 1 are: raw material A ratio, raw material B ratio, reaction temperature, and reaction time, respectively: 50%, 30%, 30℃, and 60min. The historical configuration data for four parameters in production batch 1 are 55%, 28%, 32℃, and 65min, respectively. The historical configuration data are normalized using a minimum-maximum normalization method. For example, the minimum and maximum values ​​for the proportion of raw material A are 30% and 70%, the minimum and maximum values ​​for the proportion of raw material B are 10% and 50%, the minimum and maximum values ​​for the reaction temperature are 10℃ and 50℃, and the minimum and maximum values ​​for the reaction time are 40min and 80min. Therefore, the normalized values ​​for the historical configuration data of the proportion of raw material A, the proportion of raw material B, the reaction temperature, and the reaction time are as follows: , , , In production batch 1, the degree of matching between each configuration parameter and the standard value was 0.81, -0.55, 0.54, and 0.7, respectively.

[0035] The estimated value V1 for production batch 1 is calculated using the first formula. This estimated value is a quantitative indicator of the performance or quality of the household disinfectant and can be used to assess whether the production batch meets the expected quality standards. The performance or quality of the disinfectant may be determined by multiple key indicators, such as sterilization rate, pH value, viscosity, and irritation. Setting an overall quantitative indicator can integrate these different dimensions of indicators to provide a comprehensive assessment, identifying configuration parameters with a significant impact on the target variable from production batches that meet quality standards. Defining it as the degree of contribution is to ensure that the degree of contribution is non-negative, even if The value is the opposite of the standard value (i.e. Even if the value is less than the standard value, its contribution is still positive. This reflects the degree to which the configuration parameter deviates from the standard value, rather than the direction of deviation, thus more accurately identifying key parameters.

[0036] V1= + + The values ​​of 0.3375, 0.165, 0.18, and 0.2333 represent the contributions of the proportions of raw material A, raw material B, reaction temperature, and reaction time to the quality of the household disinfectant, respectively. If the value is greater than the preset threshold (assumed to be 0.9), it means that the production batch is relatively close to the standard batch. Therefore, the corresponding key parameters can be obtained based on production batch 1. The contribution of raw material A ratio, reaction time, raw material B ratio and reaction temperature decreases in that order. Therefore, the configuration parameters (raw material A ratio, reaction time, raw material B ratio) greater than 0.17 can be set as the key parameters of production batch 1. These key parameters are configuration parameters that deviate from the standard value and need to be analyzed.

[0037] In one specific embodiment, calculating the degree of matching between historical configuration data and standard values ​​for each configuration parameter specifically includes the following steps: The matching degree between the historical configuration data and the standard value of the k-th configuration parameter in the i-th production batch is calculated based on the second formula. The second formula is: ,in, This represents the actual value of the k-th configuration parameter in the i-th production batch. This is the standard value for the kth configuration parameter.

[0038] Specifically, the matching degree of raw material A ratio, raw material B ratio, reaction temperature, and reaction time in production batch 1 is as follows: , , , .

[0039] In one specific embodiment, the performance prediction model outputs the batch type for each production batch, which includes the following steps: The performance prediction model is a regression model. The regression model generates predicted values ​​for each key indicator, sets a standard range for each key indicator, and determines whether each predicted value is within the standard range. If it is, the key indicator corresponding to the predicted value is defined as a reliable prediction indicator. If the total number of key indicators that are reliable prediction indicators in a production batch is greater than or equal to the third threshold, the production batch is defined as a normal batch; otherwise, it is defined as an abnormal batch.

[0040] Specifically, assuming there are three key indicators: sterilization rate, pH value, and viscosity, with standard ranges of 99%-100%, 6-8, and 10-50 mPa·s respectively, a trained performance prediction model is used to generate predicted values ​​for each key indicator, assuming they are 98%, 6, and 20 mPa·s respectively. Since these values ​​are all within the standard range, all key indicators are reliable predictors, with a total of 2. A third threshold is set to 3 (the total number of key indicators). Because the total number of indicators is 2, which is less than the third threshold of 3, the current batch is considered an abnormal batch. By using a regression model to predict key indicators, abnormal batches can be quickly identified, and corrective measures can be taken in a timely manner to prevent unqualified products from entering the market. At the same time, the subjectivity and error of manual judgment are reduced, and the automation and standardization of quality control are achieved.

[0041] In one specific embodiment, dangerous parameters are obtained from key parameters based on the cause of the anomaly, and the cause of the anomaly is obtained based on the dangerous parameters, which includes the following steps: Collect simulated configuration data of key parameters corresponding to all causes of anomalies, extract multiple feature quantities from the simulated configuration data, and define the causes of anomalies and feature quantities as a second dataset; Create a first random forest model and input all the second datasets into the first random forest model for training. The first random forest model calculates the change in Gini coefficient before and after splitting for each feature in each decision tree. Taking the root node of each decision tree as the starting node, determine whether there is a path in the decision tree with the feature as the splitting node. If so, define the decision tree as the first decision tree. Calculate the sum of the changes in Gini coefficients corresponding to all first decision trees and the total number of first decision trees. Define the ratio of the sum of the changes in Gini coefficients to the total number of trees as the risk index of the feature. Arrange the risk indices of all features corresponding to the cause of the anomaly in order. Define the key parameters corresponding to the features ranked above the fourth threshold as risk parameters. Create a second random forest model. Input the simulated configuration data corresponding to the hazard parameters and the corresponding anomaly causes into the second random forest model. The random forest model outputs the corresponding anomaly causes. Input the real-time configuration data into the second random forest model. The second random forest model outputs the anomaly causes for the current batch.

[0042] Specifically, during the production process, simulated anomalies are introduced. For each predefined anomaly cause, such as anomaly cause A (raw material contamination: possibly caused by impurities in the raw materials themselves or improper storage conditions), the corresponding key parameter data is: raw material A ratio = 55%, raw material B ratio = 28%, temperature = 30°C, and reaction time = 65 minutes; anomaly cause B (chemical reaction anomaly: possibly caused by improper control of reaction temperature, pH value, or reaction time), the corresponding key parameter data is: raw material A ratio = 45%, raw material B ratio = 32%, temperature = 40°C, and reaction time = 50 minutes. Feature quantities are extracted from the simulated configuration data, such as: raw material A ratio, raw material B ratio, reaction temperature, and reaction time.

[0043] The first random forest model is a general model used to analyze and evaluate the importance of features, and to identify hazardous parameters from the features. Hazardous parameters refer to those key parameters that have the greatest impact on the causes of anomalies.

[0044] Random forests consist of multiple decision trees, each of which is an independent tree-structured model. The change in the Gini coefficient represents the reduction in dataset impurity after splitting at a certain feature. The larger the change, the greater the contribution of that feature to the model's prediction. If feature A is used as a split node, it means that feature A contributes to the model's prediction. Therefore, the sum of the changes in the Gini coefficients of all first decision trees (decision trees with feature A as a split node) and the total number of first decision trees are calculated, representing the total contribution of feature A in all decision trees and the number of decision trees with feature A as a split node, respectively.

[0045] If the risk index is defined solely by the sum of the changes in the Gini coefficient, the importance of a feature will directly depend on how many decision trees it is used in. Some features may be used in more decision trees due to the structure of the random forest model or the characteristics of the training data, but this does not mean that they contribute more to each decision tree. This invention uses the ratio of the sum of the changes in the Gini coefficient to the total number of trees (average change in the Gini coefficient) as the risk index for each feature, which allows for a fair comparison of the importance of different features, even if they are used at different frequencies in the decision trees. For example, feature A is used in 5 decision trees, with a total Gini coefficient change of 0.5 and a risk coefficient of 0.5 / 5 = 0.1; feature B is used in 10 decision trees, with a total Gini coefficient change of 0.8 and a risk coefficient of 0.8 / 10 = 0.08. If only the sum is used, feature B (0.8) will be higher than feature A (0.5), but this does not truly reflect their importance. By dividing by the total number of trees, the average value of feature A (0.1) is higher than that of feature B (0.08), which more accurately reflects that feature A has a greater impact on the model.

[0046] Then, based on the above method, the risk index of all feature quantities corresponding to the abnormal cause is arranged in order, and the key parameter corresponding to the feature quantity whose ranking is greater than the fourth threshold (which can be set to 3) is defined as the risk parameter.

[0047] The main objective of creating the second random forest model is to use hazard parameters to predict the specific causes of anomalies in the current production batch. Assuming the hazard parameters for anomaly cause A (raw material contamination) are a raw material A proportion of 55% and a reaction time of 65 minutes, each simulated configuration is associated with a specific anomaly cause. The simulated configuration data corresponding to the hazard parameters and the corresponding anomaly causes are input into the second random forest model for training. The model learns the changing patterns of hazard parameters under different anomalies, thus enabling it to predict the anomaly cause corresponding to new input data. Real-time monitoring data of the current production batch is input into the trained second random forest model. Based on the learned patterns, the model predicts the anomaly cause for the current batch. For example, if the input real-time monitoring data is a raw material A proportion of 54% and a reaction time of 66 minutes, the predicted anomaly cause is likely raw material contamination.

[0048] In one specific embodiment, iterative optimization of the hazardous parameters includes the following steps: Multiple hazard parameters are combined to generate a state vector. The adjusted value is defined as the adjustment amount of the hazard parameters. The state vector is used as the state input of the deep reinforcement learning model, and the adjustment amount of the hazard parameters is used as the action output of the deep reinforcement learning model. The reward function of the deep learning model is set as follows: ,in The adjustment amount of the hazard parameter taken under state vector s The reward value, where m is the total number of key metrics. Let be the weight parameter for the i-th key indicator. The predicted values ​​of key indicators obtained by the performance prediction model are based on the state vector s and the adjustment amount of the danger parameter is q. Let i be the standard value of the i-th key indicator. To adjust the coefficients, the adjustment amount of the dangerous parameters in each abnormal batch is dynamically adjusted within a preset range until the reward function reaches its maximum value.

[0049] Specifically, identify the hazardous parameters that have the greatest impact on the production process, and combine the values ​​of these hazardous parameters into a vector, for example, [Raw Material A Ratio, Reaction Time] = [54%, 66]. Based on experience and experimental data, set the adjustment range for each hazardous parameter, for example, the adjustment amount for the Raw Material A Ratio is ± %, Adjustment amount for reaction time: ± min takes the state vector as the input state of the deep reinforcement learning model, outputs the adjustment amount of each danger parameter, and defines a reward function to evaluate the adjustment effect.

[0050] The first term in the reward function of this invention: This represents the deviation term, used to assess the deviation between the predicted value and the standard value of a key indicator. The predicted value is obtained by inputting the adjusted hazard parameter along with other key parameters into the performance prediction model to obtain the predicted value of the key indicator. Weighting parameters... This reflects the importance of different key indicators. For non-toxic household disinfectants, sterilization rate and pH value are the most important key indicators, followed by viscosity and stability. Assuming key indicator 1: sterilization rate, standard value (θ1): 99%, predicted value (ρ1(s,q)): 94%, weight parameter (ω1): 0.4, sterilization rate deviation: -0.4 * (94% - 99%) = -0.4 * (-5%) = 0.02; key indicator 2: pH value, standard value (θ2): 7.0, predicted value (ρ2(s,q)): 6.5, weight parameter (ω2), pH value deviation: 0.3, = -0.3 * (6.5 - 7.0) = -0.3 * (-0.5) = 0.015. Based on the above method, the viscosity deviation and stability deviation are calculated to be 0.02 and 0.005 respectively. The total deviation is = 0.02 + 0.015 + 0.02 + 0.005 = 0.06; Second item: This represents the squared deviation term, which is the sum of the squares of the deviations between the predicted and standard values. It is used to penalize larger deviations and encourage the model to minimize large errors when adjusting parameters. To adjust the coefficients and balance the impact of prediction bias, a squared term is used to penalize larger biases, prompting the model to adjust parameters more accurately. Calculations show that the squared bias terms for sterilization rate, pH value, viscosity, and stability are -0.025, -0.025, -0.1, and -2.5, respectively. Therefore, the total squared bias term is (-0.025) + (-0.025) + (-0.1) + (-2.5) = -2.65. The total reward is 0.06 - 2.65 = -2.59. Although the bias terms are positive, the total reward is still negative due to the squared bias term. This indicates that the model needs further parameter adjustment to reduce large errors and improve prediction accuracy until the reward function reaches its maximum value. The purpose of this process is to continuously adjust the critical parameters to maximize the reward function value, i.e., minimize bias and squared bias.

[0051] In one specific embodiment, dynamically adjusting the adjustment amount of the hazardous parameter in each abnormal batch within a preset range specifically includes the following steps: Obtain the minimum standard for each key indicator, and calculate the adjustment amount of the hazard parameter at time t based on the third formula. The third formula is: ,in, This represents the preset optimal adjustment amount for the danger parameter at time t. The weighting coefficients for the optimal adjustment amount. Given the unadjusted danger parameters, the weighting coefficients for the optimal adjustment amount are obtained based on the fourth formula, which is: ,in, As the initial weights, The value of the optimal adjustment obtained based on the reinforcement learning algorithm. To obtain the value of previously adjusted amounts based on reinforcement learning algorithms, The minimum standard for each key indicator, It is a preset positive number.

[0052] Specifically, a minimum acceptable standard is determined for each key indicator for subsequent adjustments and evaluations. The third formula is used to calculate the adjustment amount q(t) of the hazard parameter at the current moment, which incorporates the current optimal adjustment amount. and previous adjustment amount The weighted average is calculated using the weighting coefficient φ(t). A balance can be found between the old and new adjustment amounts, which can take advantage of the current optimal adjustment amount without completely abandoning the previous adjustment experience, making the adjustment process smoother, avoiding drastic fluctuations in the adjustment amount, and ensuring the stability of the production process.

[0053] The fourth formula is used to calculate the weighting coefficients of the optimal adjustment. It comprehensively considers the value of the current optimal adjustment amount and the previous adjustment amount, as well as the preset standard value and constant. b(t) represents the minimum standard for each key indicator, ensuring that the adjusted parameters are not lower than the minimum requirements and guaranteeing product quality. and These represent the values ​​of the current optimal adjustment and the previous adjustment, respectively. By comparing these two items, the merits of the new and old adjustments can be evaluated, thereby allowing for a reasonable allocation of weights.

[0054] Through such calculations and settings, the third and fourth formulas can dynamically adjust hazardous parameters, optimize the production process, and improve product quality. This method combines reinforcement learning and weighted averaging techniques, enabling it to flexibly respond to various changes in the production process and providing strong support for intelligent manufacturing.

[0055] The above describes a production monitoring method for a non-toxic household disinfectant according to embodiments of this application. The following describes a production monitoring system for a non-toxic household disinfectant according to embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the production monitoring system for a non-toxic household disinfectant in this application includes: The acquisition module is used to acquire historical configuration data of household disinfectant during the production process. The historical configuration data includes a variety of configuration parameters. The combination of various configuration parameters for each production batch is defined as a parameter configuration group, and key indicators for evaluating the quality of household disinfectant are defined. The processing module is used to measure the key indicators of each production batch of household disinfectant to obtain historical measurement data. The historical measurement data and the corresponding parameter configuration group together constitute the first dataset. The first dataset is normalized and denoised. Based on the first dataset, the historical configuration data is dimensionality reduced to obtain the key parameters. The performance prediction and anomaly analysis module is used to establish performance prediction models and anomaly analysis models. It obtains real-time configuration data of key parameters of the current batch of household disinfectant during the production process and inputs it into the performance prediction model. The performance prediction model outputs the batch type of the current batch. If the current batch is an abnormal batch, it obtains the dangerous parameters from the key parameters based on the anomaly analysis model. Key parameters other than dangerous parameters are defined as ordinary parameters. The cause of the anomaly is obtained based on the dangerous parameters. The optimization module is used to build a deep reinforcement learning model, iteratively optimize the dangerous parameters, and input the optimized dangerous parameters and normal parameters into the performance prediction model until the current batch is determined to be a normal batch.

[0056] This application also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, storing instructions that, when executed on a computer, cause the computer to perform the steps of a production monitoring method for a non-toxic household disinfectant.

[0057] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0058] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0059] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A production monitoring method for a non-toxic household disinfectant, characterized in that, include: S1: Obtain historical configuration data of household disinfectant during the production process. The historical configuration data includes various configuration parameters. Define the combination of various configuration parameters for each production batch as a parameter configuration group and define the key indicators for evaluating the quality of household disinfectant. S2: Measure the key indicators of household disinfectant for each production batch to obtain historical measurement data. The historical measurement data and the corresponding parameter configuration group together constitute the first dataset. Normalize and denoise the first dataset. Based on the first dataset, reduce the dimensionality of the historical configuration data to obtain the key parameters. S3: Establish a performance prediction model and anomaly analysis model. Obtain real-time configuration data of key parameters during the production process of the current batch of household disinfectant and input it into the performance prediction model. The performance prediction model outputs the batch type of the current batch. If the current batch is an abnormal batch, then based on the anomaly analysis model, obtain the hazardous parameters from the key parameters. Key parameters other than hazardous parameters are defined as ordinary parameters. Based on the hazardous parameters, obtain the causes of the anomalies, including collecting simulated configuration data of key parameters corresponding to all causes of anomalies, extracting multiple features from the simulated configuration data, and defining the causes of anomalies and features as the second dataset. Create a first random forest model, input all the second datasets into the first random forest model for training, and calculate the change in the Gini coefficient before and after splitting for each feature in each decision tree. The root node of each decision tree is the starting node. It is determined whether a path exists in the decision tree with a feature as a splitting node. If so, the decision tree is defined as the first decision tree. The sum of the changes in the Gini coefficient corresponding to all first decision trees and the total number of first decision trees are calculated. The ratio of the sum of the changes in the Gini coefficient to the total number of trees is defined as the risk index of the feature. The risk indices of all features corresponding to the cause of anomalies are arranged in order. The key parameters corresponding to features ranked above the fourth threshold are defined as risk parameters. A second random forest model is created. The simulated configuration data corresponding to the risk parameters and the corresponding causes of anomalies are input into the second random forest model. The random forest model outputs the corresponding causes of anomalies. Real-time configuration data is input into the second random forest model, and the second random forest model outputs the causes of anomalies for the current batch. S4: Construct a deep reinforcement learning model, iteratively optimize the hazardous parameters, and input the optimized hazardous parameters and normal parameters into the performance prediction model until the current batch is determined to be a normal batch; Iterative optimization of the hazard parameters includes: combining multiple hazard parameters to generate a state vector, defining the adjustment value as the adjustment amount of the hazard parameters, using the state vector as the state input of the deep reinforcement learning model, using the adjustment amount of the hazard parameters as the action output of the deep reinforcement learning model, and setting the reward function of the deep learning model as follows: ,in Let q be the reward value for adjusting the danger parameter under state vector s, and m be the total number of key indicators. Let be the weight parameter for the i-th key indicator. The predicted values ​​of key indicators obtained by the performance prediction model are input to the state vector s, with the adjustment amount of the danger parameter being q. Let i be the standard value of the i-th key indicator. To adjust the coefficients, the adjustment amount of the dangerous parameters in each abnormal batch is dynamically adjusted within a preset range until the reward function reaches its maximum value.

2. The production monitoring method for a non-toxic household disinfectant according to claim 1, characterized in that, The performance prediction model outputs the batch type for each production batch, including: The performance prediction model is a regression model. The regression model generates predicted values ​​for each key indicator, sets a standard range for each key indicator, and determines whether each predicted value is within the standard range. If it is, the key indicator corresponding to the predicted value is defined as a reliable prediction indicator. If the total number of key indicators that are reliable prediction indicators in the production batch is greater than or equal to the third threshold, the production batch is defined as a normal batch; otherwise, it is defined as an abnormal batch.

3. A production monitoring system for a non-toxic household disinfectant, used to implement the production monitoring method for a non-toxic household disinfectant as described in claim 2, characterized in that, include: The acquisition module is used to acquire historical configuration data of household disinfectant during the production process. The historical configuration data includes a variety of configuration parameters. The combination of various configuration parameters for each production batch is defined as a parameter configuration group, and key indicators for evaluating the quality of household disinfectant are defined. The processing module is used to measure the key indicators of each production batch of household disinfectant to obtain historical measurement data. The historical measurement data and the corresponding parameter configuration group together constitute the first dataset. The first dataset is normalized and denoised. Based on the first dataset, the historical configuration data is dimensionality reduced to obtain the key parameters. The performance prediction and anomaly analysis module is used to establish performance prediction models and anomaly analysis models. It obtains real-time configuration data of key parameters of the current batch of household disinfectant during the production process and inputs it into the performance prediction model. The performance prediction model outputs the batch type of the current batch. If the current batch is an abnormal batch, it obtains the dangerous parameters from the key parameters based on the anomaly analysis model. Key parameters other than dangerous parameters are defined as ordinary parameters. The cause of the anomaly is obtained based on the dangerous parameters. The optimization module is used to build a deep reinforcement learning model, iteratively optimize the dangerous parameters, and input the optimized dangerous parameters and normal parameters into the performance prediction model until the current batch is determined to be a normal batch.

4. A computer-readable storage medium storing instructions thereon, characterized in that, When the instruction is executed by the processor, it implements the production monitoring method for a non-toxic household disinfectant as described in claim 2.

Images