Quality prediction model training method, product quality prediction method and device

Abstract

The embodiment of the invention provides a quality prediction model training method, a product quality prediction method and a product quality prediction method and device based on a cement production scenario, and the quality prediction model training method comprises the steps: obtaining historical measurement data of a key point location in a target production scenario and reference quality parameters of a product in the target production scenario, the key point locations are point locations affecting the product quality, and the reference quality parameters are obtained by performing quality detection on the product; based on the sampling time of the reference quality parameters, distribution information of target measurement data is estimated from historical measurement data through time delay analysis, and the target measurement data are measurement data influencing the reference quality parameters; inputting the distribution information of the target measurement data into a quality prediction model to obtain a quality prediction result of the product; and training a quality prediction model based on the quality prediction result and the reference quality parameters. And the robustness of the model in a complex environment and the accuracy of prediction are improved.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of data processing technology, and in particular to a quality prediction model training method and a product quality prediction method. Background Art

[0002] With the development of science and technology, the industrial production and manufacturing field has put forward higher requirements on the real-time and accuracy of product quality control.

[0003] At present, in order to ensure that the quality indicators of products are monitored during the production process, prediction models are usually used to obtain easily measurable data online to estimate quality characteristics that are difficult to measure directly. These models analyze the easily measurable data obtained online to obtain the status of key quality indicators. At the same time, due to various unforeseen changes in the production environment, such as fluctuations in raw material composition and changes in equipment health status, the constructed model will have certain errors. Therefore, the output correction method is usually used to correct the output results of the constructed model.

[0004] However, although the above method corrects the model results, due to the uncertainty of sampling time in the production process and the difference in time lag between different quality indicators, the model based on fixed parameters is prone to the situation where the prediction results are inconsistent with the actual values, resulting in low prediction accuracy and large deviation. Therefore, a more reasonable and efficient model training method is urgently needed. Summary of the invention

[0005] In view of this, an embodiment of this specification provides a quality prediction model training method. One or more embodiments of this specification also involve a product quality prediction method, a product quality prediction method based on cement production scenarios, a quality prediction model training device, a product quality prediction device, a product quality prediction device based on cement production scenarios, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects existing in the prior art.

[0006] According to a first aspect of an embodiment of this specification, a quality prediction model training method is provided, including: Obtain historical measurement data of key points in the target production scenario and reference quality parameters of products in the target production scenario, where key points are points that affect product quality and reference quality parameters are obtained by quality testing of products; Based on the sampling time of the reference quality parameter, the distribution information of the target measurement data is estimated from the historical measurement data through time lag analysis, wherein the target measurement data is the measurement data that affects the reference quality parameter; Input the distribution information of the target measurement data into the quality prediction model to obtain the quality prediction result of the product; The quality prediction model is trained based on the quality prediction results and reference quality parameters.

[0007] According to a second aspect of the embodiments of this specification, a product quality prediction method is provided, comprising: Obtain current measurement data of key points in the target production scenario; The current measurement data is input into the quality prediction model to obtain the quality prediction result of the product under the target production scenario, wherein the quality prediction model is trained according to the above method.

[0008] According to a third aspect of the embodiments of this specification, a product quality prediction method based on a cement production scenario is provided, comprising: Obtain current measurement data of key points in cement production scenarios; The current measurement data is input into the quality prediction model to obtain a quality prediction result for the free calcium content of cement clinker, wherein the quality prediction model is trained according to the above method.

[0009] According to a fourth aspect of the embodiments of this specification, a quality prediction model training device is provided, including: A first acquisition module is configured to acquire historical measurement data of key points in a target production scenario and reference quality parameters of products in the target production scenario, wherein the key points are points that affect product quality, and the reference quality parameters are obtained by quality testing of the products; an estimation module, configured to estimate distribution information of target measurement data from historical measurement data through time lag analysis based on a sampling time of the reference quality parameter, wherein the target measurement data is measurement data that affects the reference quality parameter; A first prediction module is configured to input the distribution information of the target measurement data into the quality prediction model to obtain a quality prediction result of the product; The training module is configured to train the quality prediction model based on the quality prediction result and the reference quality parameter.

[0010] According to a fifth aspect of the embodiments of this specification, there is provided a product quality prediction device, comprising: A second acquisition module is configured to acquire current measurement data of key points in a target production scene; The second prediction module is configured to input the current measurement data into a quality prediction model to obtain a quality prediction result of the product under a target production scenario, wherein the quality prediction model is trained according to the above method.

[0011] According to a sixth aspect of the embodiments of this specification, a product quality prediction device based on a cement production scenario is provided, comprising: A third acquisition module is configured to acquire current measurement data of key points in the cement production scene; The third prediction module is configured to input the current measurement data into a quality prediction model to obtain a quality prediction result for the free calcium content of cement clinker, wherein the quality prediction model is trained according to the above method.

[0012] According to a seventh aspect of an embodiment of this specification, there is provided a computing device, including: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer program / instructions. When the computer program / instructions are executed by the processor, the steps of the above methods are implemented.

[0013] According to an eighth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores a computer program / instruction, and the program / instruction implements the steps of the above methods when executed by a processor.

[0014] According to a ninth aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program / instruction, which implements the steps of the above-mentioned methods when executed by a processor.

[0015] An embodiment of the present specification implements a quality prediction model training method, including: obtaining historical measurement data of key points in a target production scenario and reference quality parameters of products in the target production scenario, wherein the key points are points that affect product quality, and the reference quality parameters are obtained by quality testing of the products; based on the sampling time of the reference quality parameters, the distribution information of the target measurement data is estimated from the historical measurement data through time lag analysis, wherein the target measurement data is the measurement data that affects the reference quality parameters; inputting the distribution information of the target measurement data into the quality prediction model to obtain the quality prediction result of the product; and training the quality prediction model based on the quality prediction result and the reference quality parameters. By introducing time lag analysis to deal with the uncertainty between the historical measurement data of key points and the reference quality parameters of the products, the quality prediction model based on this training can more accurately capture the time dimension characteristics of the feature variables, thereby improving the robustness and accuracy of the quality prediction model, and adjusting the quality prediction model based on the quality prediction results and the actual reference quality parameters, realizing the continuous optimization of the quality prediction model, ensuring that it can still provide high-precision quality prediction in a complex and changeable industrial production environment, which helps to improve production efficiency, reduce costs and optimize resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of a quality prediction model training method provided by an embodiment of this specification; Figure 2 is a flow chart of a product quality prediction method provided by an embodiment of this specification; Figure 3 is a flowchart of a product quality prediction method based on a cement production scenario provided by an embodiment of this specification; Figure 4a It is a schematic diagram of the overall structure of a product quality prediction method provided by an embodiment of this specification; Figure 4b It is a data diagram of a kiln head fire temperature segment in the overall structural schematic diagram of a product quality prediction method provided in one embodiment of this specification; Figure 4c It is a kiln current fragment data diagram in the overall structural schematic diagram of a product quality prediction method provided by an embodiment of this specification; Figure 4d It is a distribution information data diagram of the kiln head fire temperature in the overall structural schematic diagram of a product quality prediction method provided by an embodiment of this specification; Figure 4e It is a distribution information data diagram of kiln current in the overall structural schematic diagram of a product quality prediction method provided by an embodiment of this specification; Figure 4f It is a distribution information data diagram of predicting free calcium content in an overall structural schematic diagram of a product quality prediction method provided in one embodiment of this specification; Figure 4g It is an updated model parameter distribution information data diagram in the overall structural diagram of a product quality prediction method provided by an embodiment of this specification; Figure 5 is a processing flow chart of a quality prediction model training method provided by an embodiment of this specification; Figure 6 It is a structural schematic diagram of a quality prediction model training device provided by an embodiment of this specification; Figure 7 It is a structural schematic diagram of a product quality prediction device provided by an embodiment of this specification; Figure 8 It is a structural schematic diagram of a product quality prediction device based on a cement production scenario provided by an embodiment of this specification; Fig. 9 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0017] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0018] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0019] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0020] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0021] First, the terms involved in one or more embodiments of this specification are explained.

[0022] Soft sensing: Soft sensing is a technology that uses mathematical models and statistical methods to estimate quality or performance indicators that are difficult to measure directly by using easily measurable process variables.

[0023] Linear regression model: Linear regression model (LRM or LR) is a statistical model used to establish a linear relationship between a dependent variable and one or more independent variables, and is often used in predictive analysis.

[0024] Bayes' Theorem: Bayes' Theorem is a probability theory formula that describes how the probability of an event occurring under given conditions is updated as new evidence is introduced.

[0025] Covariance Matrix: The covariance matrix (usually represented by ∑) is a matrix composed of the covariances between variables in a multidimensional data set, which is used to measure the linear correlation between variables.

[0026] Random sampling: Random sampling (RS) is a method of randomly selecting samples from a population according to a certain probability mechanism, ensuring that each individual has the same chance of being selected to reduce bias and improve sample representativeness.

[0027] At present, in order to ensure that the quality indicators of products are monitored during the production process, prediction models are usually used to obtain easily measurable data online to estimate quality characteristics that are difficult to measure directly. These models analyze the easily measurable data obtained online to obtain the status of key quality indicators. At the same time, due to various unforeseen changes in the production environment, such as fluctuations in raw material composition and changes in equipment health status, the constructed model will have certain errors. Therefore, the output correction method is usually used to correct the output results of the constructed model.

[0028] However, although the above method corrects the results of the model, due to the uncertainty of sampling time in the production process and the differences in time lags between different quality indicators, the model based on fixed parameters is prone to the situation where the predicted results are inconsistent with the actual values, resulting in low prediction accuracy and large deviations.

[0029] In view of the above problems, a quality prediction model training method is provided in this specification. One or more embodiments of this specification also involve a product quality prediction method, a product quality prediction method based on cement production scenarios, a quality prediction model training device, a product quality prediction device, a product quality prediction device based on cement production scenarios, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail one by one in the following embodiments.

[0030] See also Figure 1 , Figure 1 A flow chart of a quality prediction model training method provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0031] Step 102: Obtain historical measurement data of key points in the target production scenario and reference quality parameters of products in the target production scenario, wherein the key points are points that affect product quality, and the reference quality parameters are obtained by quality testing of the products.

[0032] The target production scenario is a specific production process or link in a specific industrial or manufacturing environment, such as the clinker production scenario in the cement production process, the coating production scenario in the automobile manufacturing process, and the drug synthesis reaction generation scenario in the pharmaceutical process. The target generation scenario usually has clear product quality requirements and associated process control requirements. There are data sources that can be collected or measured in the target production scenario, such as the measurement of physical parameters such as temperature and pressure.

[0033] Key points are points that affect product quality. Specifically, key points refer to the corresponding states of key positions or influencing parameters of specific links that can significantly affect the quality of the final product during the production process. These key points can be locations in a physical sense (such as the operating status of a certain equipment on the production line), or they can be certain parameters or variables in the production process (such as temperature, pressure, flow rate, etc.). For example, in the scenario of body welding in the automobile production process, the key point can be whether the operating status of the welding machine is normal, and the operating status of the welding machine directly affects the production quality of the body; in the scenario of clinker production in the cement production process, the key points can be parameters such as the kiln head fire temperature, the NOx concentration in the smoke chamber, the secondary air temperature, and the kiln current, which directly affect the product quality of the cement.

[0034] Quality parameters refer to indicators used to evaluate the quality characteristics of products. Quality parameters can be certain attribute values ​​of products measured directly or obtained through indirect calculation. The selection of quality parameters depends on the nature of specific products or industry standards. In practical applications, most quality parameters are parameters that are not easy to measure directly. These quality parameters can be affected by key points, so the values ​​of quality parameters can be obtained through the measurement data of key points. For example, in the cement production process, free calcium content is an important quality parameter that reflects the strength and other performance characteristics of cement products, but the free calcium content is usually not easy to measure directly during the calcination process. Therefore, the free calcium content can be reflected by the measurement data of key points such as the kiln head fire temperature, the NOx concentration in the smoke chamber, the secondary air temperature, the kiln current and other parameters.

[0035] Reference quality parameters refer to the quality parameters obtained by laboratory analysis or other precise measurement of actual products. Reference quality parameters can be used as a reference standard for product quality. For example, in the cement production process, cement product samples are collected directly from the calcining kiln and analyzed in the laboratory to obtain the actual free calcium content as a reference quality parameter.

[0036] In actual industrial production, the measurement of product quality parameters often cannot be measured directly or the measurement will affect the product itself. Therefore, it is necessary to obtain measurement data of key points that affect product quality to reflect the quality parameters of the product. This method can be called a soft measurement method. For example, in the clinker production scenario of cement production, since it is not convenient to directly measure the free calcium content of cement from the inside of the calcining kiln, continuous collection of cement samples for laboratory analysis and testing will lead to a decrease in the final product output and an increase in production costs. In addition, frequent switching of the calcining kiln will affect the original cement production process, further affecting the quality of the cement product. In the coating production scenario of the automobile manufacturing process, since directly measuring the thickness and uniformity of the automobile paint surface will cause the automobile paint surface itself to be irreversibly affected, it is necessary to obtain the parameters such as the pressure, spraying distance, paint flow rate and spray pattern of the spray gun during the paint spraying process to reflect the quality parameters of the thickness and uniformity of the automobile paint surface.

[0037] For example, taking the clinker burning scenario in the cement production process as an example, the measurement data of parameters such as the kiln head fire temperature, smoke chamber NOx concentration, secondary air temperature, kiln current, etc. in the clinker burning scenario are obtained, and the actual free calcium content is obtained as a reference quality parameter by obtaining cement product samples for laboratory analysis.

[0038] In this step, by obtaining historical measurement data of key points in the target production scenario and reference quality parameters of the products in the target production scenario, a data basis is provided for the subsequent training of the quality prediction model.

[0039] Step 104: Based on the sampling time of the reference quality parameter, distribution information of the target measurement data is estimated from the historical measurement data through time lag analysis, wherein the target measurement data is the measurement data that affects the reference quality parameter.

[0040] Sampling time refers to the specific moment when the product sample corresponding to the reference quality parameter is sampled. In industrial production, reference quality parameters are usually obtained through laboratory analysis or other precise measurement methods. Different sampling times reflect the sampling time points and the order of multiple samples. For example, in the clinker burning scenario in the cement production process, in order to obtain the reference quality parameter as a reference, that is, the free calcium content of the cement product, it is necessary to collect samples of the cement product from the calcining kiln and send them to the laboratory for analysis and testing of the free calcium content. The moment when the cement product sample is collected is the sampling time. The reference quality parameters at the sampling time can reflect the actual quality status of the product at the sampling time.

[0041] The target measurement data is the specific measurement data estimated from the historical measurement data according to the time lag analysis. The target measurement data is the measurement data that affects the reference quality parameter. The distribution information of the target measurement data is the distribution of the target measurement data in the time dimension, which reflects the probability of the target measurement data affecting the reference quality parameter.

[0042] Time lag analysis refers to the process of analyzing the target measurement data corresponding to the reference quality parameters that affect the sampling time. Since the production process of a product cannot be completed in an instant, the reference quality parameters of the product at the sampling time are generally affected by the target measurement data of the key points corresponding to a previous target moment, that is, the impact of the measurement data of certain input key points will not be immediately reflected in the output quality parameters, but will only appear after a period of time, reflecting the time lag relationship between the reference quality parameters and the target measurement data. For example, in the clinker burning scenario in the cement production process, the current kiln head fire temperature and kiln current parameters may affect the quality of cement clinker half an hour or even longer later. Time lag analysis aims to estimate the target measurement data corresponding to the reference quality parameters at the sampling time from all historical measurement data based on this time lag relationship, so as to ensure that the data used for subsequent quality prediction model training can accurately correspond to the actual quality changes.

[0043] The distribution information describes the probability distribution of possible values ​​of a variable or a group of variables, and describes the frequency or probability of occurrence of this variable or this group of variables under different values. The distribution information can be represented by a variety of statistics and graphs, including mean, variance, standard deviation, and possible skewness and kurtosis. In the embodiments of this specification, the distribution reflects the change pattern and trend of the target measurement data.

[0044] A time-lag analysis is performed based on the sampling time of the reference quality parameters to determine the target measurement data of the key points of the reference quality that affect the sampling time. The target measurement data is analyzed to estimate the distribution of the target measurement data, reflecting the change pattern and trend of the target measurement data of the reference quality parameters that affect the sampling time when the time-lag relationship exists.

[0045] For example, taking the clinker burning scenario in the cement production process as an example, when the free calcium content, that is, the reference quality parameter, is obtained from the laboratory test of a batch of product samples, the sampling time of this batch of product samples, that is, the sampling time point, can be used in the previously recorded historical measurement data, that is, the historical measurement values ​​of the kiln head fire temperature, kiln current and other data. Through time-lag analysis, the target measurement data such as the kiln head fire temperature and kiln current that affect the free calcium content of the sampling time can be found, and the distribution information of these target measurement data can be estimated.

[0046] In this step, by performing time lag analysis based on the sampling time of the reference quality parameters, the distribution information of the target measurement data is estimated from the historical measurement data of key points, thereby providing an accurate data basis for the subsequent quality prediction model training and ensuring that the quality prediction model can accurately reflect the actual situation in the production process.

[0047] Step 106: Input the distribution information of the target measurement data into the quality prediction model to obtain the quality prediction result of the product.

[0048] A quality prediction model is a mathematical or statistical model that analyzes and learns the relationship between historical measurement data and product quality parameters at key points in the target production scenario, and predicts product quality based on the input measurement data. The quality prediction model can use easily accessible measurement data (such as temperature, pressure, etc.) to estimate quality parameters that are difficult to measure directly (such as free calcium content in cement), thereby achieving real-time monitoring and optimization of product quality.

[0049] The quality prediction model can be of various types, and the specific choice depends on the needs of the application field and the technical maturity. Common models can be linear regression models, nonlinear regression models, neural network models, etc. Quality prediction models are usually trained based on historical data, that is, past key point measurement data and corresponding laboratory measured quality parameters can be used as training sets, and the mapping relationship between the two can be found through machine learning algorithms.

[0050] The quality prediction result refers to the distribution of product quality characteristic values ​​calculated by the quality prediction model based on the input target measurement data distribution information. The quality prediction result reflects the distribution of the degree of influence of the input target measurement data on the quality characteristics of the produced products under the conditions of the target production scenario. For example, in the clinker burning scenario of the cement production process, the quality prediction result can be the distribution of the predicted value of the free calcium content; in the coating production scenario of the automobile manufacturing process, the quality prediction result can be the distribution of the predicted value of the paint thickness or uniformity.

[0051] The distribution information of the target measurement data is input into the quality prediction model to obtain the quality prediction result. Specifically, the distribution information of the target measurement data at key points estimated after time lag analysis is used as the input of the model, and the quality prediction model is used for reasoning to predict the corresponding product quality distribution.

[0052] Due to the existence of a time lag relationship, and the time lag relationship can be affected by various factors such as raw material differences and equipment conditions in the actual production process, the lag time is not a constant value, that is, the target measurement data obtained by time lag analysis based on the sampling time at a certain moment is not a constant value, but presents a certain distribution within a time range. Therefore, when using the quality prediction model to predict quality based on the target measurement data, the entire distribution information of the target measurement data needs to be used as the input of the model so that the uncertainty in the time dimension can be fully considered.

[0053] For example, taking the clinker burning scenario in the cement production process as an example, after obtaining the distribution information of target measurement data such as the kiln head fire temperature and kiln current that affect the reference free calcium content at the sampling time, these distribution information can be input into the quality prediction model to obtain the predicted free calcium content at the sampling time predicted by the model.

[0054] In this step, the distribution information of the target measurement data of the key points estimated after time-lag analysis is input into the quality prediction model to obtain the quality prediction result of the product, taking into account the uncertainty in the time dimension. At the same time, the quality prediction result also reflects the distribution of the prediction result of the current quality prediction model, providing a data basis for the subsequent training of the quality prediction model in combination with the reference quality parameters.

[0055] Step 108: Based on the quality prediction result and the reference quality parameter, the quality prediction model is trained.

[0056] The quality prediction result is the distribution of predicted values ​​obtained by the quality prediction model based on the distribution information of the target measurement data; the reference quality parameter is the actual value obtained by analyzing and measuring the product samples actually affected by the distribution information of the target measurement data.

[0057] The error information is determined based on the difference or error value between the quality prediction result and the reference quality parameter, and the model parameters are actually updated based on the error information to make the quality prediction result more consistent with the reference quality parameter. That is, the prediction quality model is trained to improve the prediction accuracy of the prediction quality model.

[0058] After completing a training of the quality prediction model, additional data sets can be used to verify the performance of the quality prediction model to eliminate the occurrence of overfitting and other phenomena, so that the accuracy of the quality prediction model can be improved and more in line with the actual production process.

[0059] After obtaining the quality prediction model with the target accuracy, that is, completing the training of the quality prediction model, the quality prediction model can correctly obtain the quality parameters of the product based on the input key point measurement data, thereby avoiding the need to obtain the product quality parameters by measuring the product through laboratory analysis or other high-cost measurement methods. This provides more reliable technical support for product quality control in industrial production and ensures the consistency and reliability of product quality.

[0060] In the embodiments of the present specification, by obtaining historical measurement data and reference quality parameters of key points in the target production scenario, and introducing time lag analysis to deal with the uncertainty between the historical measurement data of key points and the reference quality parameters of the product, the quality prediction model based on this training can more accurately capture the time dimension characteristics of the feature variables, thereby improving the robustness and accuracy of the quality prediction model. At the same time, the quality prediction model is adjusted based on the quality prediction results and the actual reference quality parameters, thereby achieving continuous optimization of the quality prediction model, ensuring that it can still provide high-precision quality predictions in complex and changeable industrial production environments, which helps to improve production efficiency, reduce costs and optimize resource allocation.

[0061] In an optional embodiment of the present specification, based on the sampling time of the reference quality parameter, the distribution information of the target measurement data is estimated from the historical measurement data through time lag analysis, including: Based on the sampling time of the reference quality parameter, the target time information of the product quality affecting the sampling time is determined through time lag analysis; According to the target time information, determining the target measurement data corresponding to the target time information from the historical measurement data; According to the preset time distribution characteristics, the target measurement data is analyzed to obtain the distribution information of the target measurement data.

[0062] The target time information refers to the specific time point or time period corresponding to the target measurement data of the reference quality parameter that affects the sampling time in the historical measurement data of the key points. In the historical measurement data of the key points, there may be different measurement values ​​of multiple groups of measurement data for a certain time point or a certain time period, and the distribution information of the target measurement data corresponding to the target time information can be determined based on these different measurement values. The target time information reflects the state at a certain moment in the entire production process, and this state will affect the product quality of subsequent products after a period of time, that is, according to the time lag effect.

[0063] In practical applications, the target time information is determined based on the sampling time through time lag analysis, and can be determined based on historical production processes. For example, in the cement production process, the reference quality parameters of the sampling time are often affected by the measurement data of the key points about 30 minutes ago, and the specific impact time range will also change. For example, in the initial stage of production, the temperature in the calcining kiln is low, and the impact time range is about 10 minutes at this time, then the determined target time information is about 35 minutes ago to about 25 minutes ago; in the final stage of production, the temperature in the calcining kiln is high, and the impact time range is about 2 minutes at this time, then the determined target time information is about 31 minutes ago to about 29 minutes ago. The specific time lag relationship of the target time information is determined according to different target production scenarios and different production stages, and the embodiments of this specification do not make specific limitations on this.

[0064] After the target time information is determined, the corresponding target measurement data can be determined from the historical measurement data according to the target time information. For example, when the target time information is determined to be about 31 minutes ago to about 29 minutes ago, the corresponding target measurement data is determined to be the measurement data within 2 minutes between about 31 minutes ago and about 29 minutes ago.

[0065] Time distribution characteristics refer to predefined statistical characteristics on the time dimension, such as mean, variance, skewness, kurtosis, etc., which are used to describe the change pattern and trend of the target measurement data over the entire time range. Time distribution characteristics can be determined according to different target production scenarios and different production stages, such as Gaussian distribution, Poisson distribution, uniform distribution, etc., and this specification embodiment does not make specific limitations on this.

[0066] After determining the target measurement data, the target measurement data can be analyzed according to the preset time distribution characteristics to obtain the distribution information of the target measurement data. The distribution information of the target measurement data is the distribution in the time dimension, which reflects the uncertainty in the time dimension. That is, the actual measurement value of the measurement data at each moment is known, but the specific target time information is a rough range, which presents a distribution situation to the outside.

[0067] Optionally, taking the time distribution feature of Gaussian distribution as an example, the sampling time of obtaining the reference quality parameter is time k, and the time distribution feature can be shown as formula (1): (1) The first Target measurement data The distribution information of can be shown as formula (2): (2) Among them, the mean It can be shown as formula (3): (3) variance It can be shown as formula (4): (4) For example, taking the clinker burning scene in the cement production process as an example, a sample of cement product is obtained at time k, and the reference free calcium content is obtained through laboratory testing. According to empirical data, in the final stage of production, the impact time range is about 2 minutes, then the corresponding target time information is determined to be about 31 minutes ago to about 29 minutes ago, and the corresponding target measurement data of the kiln head fire temperature is the measurement data within 2 minutes between about 31 minutes ago and about 29 minutes ago. Then, according to the target measurement data of the kiln head fire temperature, the distribution information of the kiln head fire temperature on the target time information is obtained. The distribution information can be used as the input of the quality prediction model for prediction to obtain the predicted free calcium content.

[0068] In the embodiments of this specification, through time lag analysis, the target measurement data that affects the product quality in the historical measurement data is accurately determined, and its distribution information is used as input for the quality prediction model according to the time distribution characteristics. Not only the time lag effect in the production process is taken into account, ensuring that the quality prediction model can accurately capture the impact of key points on the final product quality, but also the change pattern and trend of the target measurement data are comprehensively described through the preset time distribution characteristics, which significantly improves the accuracy and robustness of quality prediction, optimizes the quality control process in industrial production, reduces costs and improves efficiency, and ensures the consistency and reliability of product quality.

[0069] In an optional embodiment of the present specification, the target production scene includes at least two key points; according to the target time information, the target measurement data corresponding to the target time information is determined from the historical measurement data, including: According to the target time information, determine the target measurement data of each key point corresponding to the target time information from the historical measurement data; According to the preset time distribution characteristics, the target measurement data is analyzed to obtain the distribution information of the target measurement data, including: According to the preset time distribution characteristics, the target measurement data of each key point is analyzed respectively to obtain the distribution information of the target measurement data of each key point; The distribution information of the target measurement data of each key point is merged to obtain the distribution information of the target measurement data.

[0070] In the actual industrial production process, there are often more than one parameter that affects product quality. Therefore, the target production scenario often includes at least two key points.

[0071] When the target measurement data is determined from the historical measurement data according to the target time information, the target measurement data of each key point is determined.

[0072] After determining the target measurement data of each key point, each target measurement data is analyzed according to the preset time distribution characteristics to obtain the distribution information of each target measurement data.

[0073] After determining the distribution information of the target measurement data from different key points, since the target time information and time distribution characteristics are the same, these distribution information can be merged to obtain the distribution information of the target measurement data, and the distribution information of the target measurement data can be joint distribution information. The specific merging method can be determined according to different target production scenarios and different production stages, such as using covariance matrix, multivariate distribution, conditional distribution, etc., and the embodiments of this specification do not make specific limitations on this.

[0074] Optionally, taking the covariance matrix as an example, the obtained distribution information of each target measurement data is merged. Distribution information , and the joint distribution information obtained can be shown as formula (5): (5) Among them, the mean It can be shown as formula (6): (6) Among them, the covariance matrix can be shown as formula (7): (7) For example, taking the clinker burning scene in the cement production process as an example, according to the target time information of about 31 minutes ago to about 29 minutes ago, the corresponding target measurement data of the kiln head fire temperature and the kiln current are determined to be the measurement data within 2 minutes between about 31 minutes ago and about 29 minutes ago. Then, the target measurement data of the kiln head fire temperature and the target measurement data of the kiln current are analyzed according to the time distribution characteristics, and the distribution information of the kiln head fire temperature and the kiln current on the target time information is obtained. Then, the distribution information of the kiln head fire temperature and the kiln current is merged using the covariance matrix to obtain the joint distribution information, which can be used as the input of the quality prediction model for prediction to obtain the predicted free calcium content.

[0075] In the embodiments of this specification, by performing time distribution feature analysis on data at different key points respectively, and then merging these distribution information into a joint distribution as the input of the quality prediction model, the prediction accuracy and robustness of the quality prediction model are significantly improved, the accuracy and reliability of quality prediction are enhanced, the quality control process in industrial production is optimized, costs are reduced and efficiency is improved, and the consistency and reliability of product quality are ensured.

[0076] In an optional embodiment of the present specification, inputting the distribution information of the target measurement data into the quality prediction model to obtain the quality prediction result of the product includes: Inputting the distribution information of the target measurement data into the quality prediction model to obtain the predicted quality distribution corresponding to the sampling time; Based on the quality prediction results and reference quality parameters, the quality prediction model is trained, including: Based on the predicted mass distribution and the reference mass parameters, uncertainty correction amounts and estimated value correction amounts of the model parameters of the mass prediction model are calculated; Model parameters of the quality prediction model are updated according to the uncertainty correction amount and the estimate correction amount.

[0077] The quality prediction model is used to determine the quality prediction result of the product, and its input is the distribution information of the target measurement data.

[0078] The distribution information of the target measurement data is the distribution in the time dimension. For each specific moment, the corresponding definite value of the key point can be found in the historical measurement data, such as the kiln head fire temperature value or kiln current value at a certain moment. However, due to the existence of the time lag effect, the moment corresponding to the parameters of the key points affecting product quality is not a definite value, but a certain distribution information presented within a time range, and the corresponding target measurement data is also a certain distribution information. The distribution information of the input target measurement data reflects the change pattern and trend of each key point over a period of time, capturing the uncertainty in the time dimension.

[0079] The uncertainty correction is a correction factor calculated based on the distribution information of the quality prediction results and the reference quality parameters. The uncertainty correction reflects and adjusts the model's estimate of the uncertainty of the quality prediction results. The uncertainty correction helps the quality prediction model to more accurately quantify the confidence interval or error range between the quality prediction results and the reference quality parameters based on the uncertainty parameters (such as the covariance matrix) included in the distribution information of the quality prediction results.

[0080] The estimated value correction is also a correction factor calculated based on the distribution information of the quality prediction result and the reference quality parameter. The estimated value correction is used to optimize the quality prediction model's estimate of the expected output. The estimated value correction assists the quality prediction model in obtaining the output value of the quality prediction result more accurately based on the estimated value parameters (such as the mean vector) included in the distribution information of the quality prediction result, thereby improving the overall performance and accuracy of the quality prediction model.

[0081] After the quality prediction model receives the distribution information of the target measurement data as input, since the distribution information presents the probability of different values ​​to the outside, different values ​​are determined as input according to different probabilities, and the obtained results, that is, the quality prediction results are also different, that is, the quality prediction results are also presented as a corresponding distribution information. The moment of determining the target measurement data is the sampling time of the reference quality parameter, and the moment corresponding to the quality prediction result determined according to the distribution information of the target measurement data is a certain value, that is, the sampling time of the reference quality parameter. In other words, the reference quality parameter and the quality prediction result correspond to the same time, which is the sampling time.

[0082] The quality prediction result is determined in the time dimension, so the distribution information presented by the quality prediction result refers to the distribution of the uncertain parameter value range at a certain moment.

[0083] That is, the input to the quality prediction model is the target measurement data distribution information which is a constant at each moment but uncertain in the time dimension, and the output of the quality prediction model is the quality prediction result of the uncertain parameter value at the sampling time determined in the time dimension.

[0084] Optionally, for the joint distribution information The quality prediction model as input and the corresponding output quality prediction result can be expressed as formula (8): (8) For example, taking the clinker burning scenario in the cement production process as an example, the distribution information of the target measurement data input into the quality prediction model is the distribution of the measurement data of the kiln head fire temperature within 2 minutes between about 31 minutes ago and about 29 minutes ago. The output of the corresponding quality prediction model is the distribution information of the predicted free calcium content based on the distribution of parameter values ​​at the current moment.

[0085] In the embodiments of the present specification, by using the target measurement data distribution information which is a constant value at each moment but uncertain in the time dimension as the input of the quality prediction model, the distribution information of the quality prediction result which is determined in the time dimension but the parameter value is uncertain is obtained, thereby realizing the transformation from historical measurement data to product quality prediction, improving the prediction accuracy, enhancing the robustness of the model, and providing more scientific technical support for quality control in industrial production.

[0086] In an optional embodiment of the present specification, before inputting the distribution information of the target measurement data into the quality prediction model to obtain the quality prediction result of the product, the method further includes: Acquire a training sample set, wherein the training sample set includes a sample quality parameter and sample measurement data, and the sample measurement data is obtained based on a time series matching with the sample quality parameter; Using the training sample set, a quality prediction model is constructed.

[0087] A training sample set refers to a set of data containing known results (such as sample quality parameters) and their corresponding inputs (such as sample measurement data). These data are used to train the quality prediction model so that it can learn the relationship between input and output.

[0088] Sample quality parameters refer to actual product quality parameters obtained through laboratory analysis or other high-precision measurement methods. For example, in the cement production process, the sample quality parameter can be the free calcium content of a batch of cement clinker.

[0089] Sample measurement data refers to the historical measurement data of key points in the production process, which correspond to the sample quality parameters after time series matching. For example, in the cement production process, sample measurement data may include the measurement data of parameters such as the kiln head fire temperature and kiln current during the calcination of a batch of cement clinker. These measurement data are obtained by time series matching based on the free calcium content of the batch of cement clinker.

[0090] Specifically, due to the time lag effect in the production process, the key point data at a certain moment may affect the product quality at a certain point in the future. Therefore, it is necessary to determine the corresponding matching time window between the two based on the time lag analysis, so as to ensure that the quality parameters of each sample in the training sample set and its corresponding measurement data are matched in time. For example, in the cement production process, assuming that the free calcium content of a batch of clinker is detected at a certain point in time, it is necessary to trace back to a period of time before that time point to find the key point measurement data that has a significant impact on the final free calcium content and use it as the sample measurement data.

[0091] When a training sample set is obtained, the training sample set can be used to construct a quality prediction model. Specifically, when constructing a model, different types of models can be selected according to different target production scenarios and different production stages, such as linear regression models, nonlinear regression models, neural network models, etc., and this specification embodiment does not make specific limitations on this.

[0092] After determining the type of quality prediction model, the data of the training sample set can be cleaned and filtered, and then the model can be trained using machine learning or deep learning methods to complete the construction of the quality prediction model.

[0093] For example, taking the clinker burning scenario in the cement production process as an example, after obtaining the free calcium content of a batch of cement clinker, the measurement data of parameters such as the kiln head fire temperature and kiln current during the calcination process of the batch of cement clinker can be determined by time series matching based on the free calcium content; the free calcium content, kiln head fire temperature, current and other parameters are used as training sample sets to construct a quality prediction model for predicting the free calcium content.

[0094] In the embodiments of this specification, by performing correlation analysis on sample measurement data and sample quality parameters after time series matching, it is ensured that the model can learn the complex relationship between key point data and final product quality, overcome the time lag effect in the production process, and improve the adaptability of the quality prediction model to multivariable systems, thereby reducing dependence on high-cost laboratory analysis, reducing production costs, and improving production efficiency.

[0095] In an optional embodiment of the present specification, a quality prediction model is constructed using a training sample set, including: Based on the training sample set, sample quality parameters and sample measurement data are randomly sampled to obtain a sampled target sample set; Using the target sample set, the initial quality prediction model is regressed to determine the model parameters of the quality prediction model; Based on the model parameters, a quality prediction model is constructed.

[0096] After obtaining the training sample set including sample quality parameters and sample measurement data, the quality prediction model can be constructed. Specifically, the training sample set can be randomly sampled to obtain the sampled target sample set. Random sampling refers to randomly selecting a part of the data from the original training sample set as the sampled target sample set. Its purpose is to: adjust the appropriate amount of data to avoid too many training sample sets affecting the training efficiency; help ensure that the model does not overfit a specific data model and improve the generalization of the model; and can also train multiple quality prediction models by obtaining multiple sampled training sample sets through random sampling, and determine the higher performance model as the constructed quality prediction model, thereby improving the robustness and stability of the model.

[0097] Regression processing is a machine learning method that aims to establish a mathematical relationship between input variables (such as sample measurement data) and output variables (such as sample quality parameters). Through regression processing, the parameter representation of the model can be obtained, and the model parameters can be trained based on the sampled training sample set to complete the model construction.

[0098] Model parameters are key values ​​within the model that determine its behavior, and usually include weights, bias terms, hidden layer parameters, and other parameters.

[0099] Exemplarily, the reference quality parameter obtained is , the sample measurement data obtained based on time series matching is .

[0100] The reference quality parameter and sample measurement data The training sample set is formed. Randomly sample the training sample set and record The target sample set obtained by sampling is The linear regression model is used to construct the quality prediction model. The structure of the constructed quality prediction model can be shown as formula (9): (9) The quality prediction model is subjected to regression processing, and the model parameters of the quality model can be obtained as shown in formula (10): (10) Since the quality prediction model trained by the sampled target sample set is not the most ideal model, the parameters of the model are not fixed, but are distributed according to the training sample set. The specific distribution can be shown as formula (11)-formula (13): (11) (12) (13) in , .

[0101] In the embodiments of this specification, random sampling is used to determine the target sample set to be adopted, which reduces the computational complexity and prevents overfitting, ensures that the model can adapt to a wider range of data patterns, performs regression processing on the initial quality prediction model, optimizes the model parameters, improves the prediction accuracy, improves the training efficiency and generalization ability of the model, enhances the robustness and stability of the model, and thus optimizes the production process, reduces production costs, and ensures the consistency and reliability of product quality.

[0102] In an optional embodiment of the present specification, after constructing the quality prediction model using the training sample set, the method further includes: Using the quality prediction model, the noise of the sample quality parameters and sample measurement data in the training sample set is estimated to obtain the noise impact factor; Based on the quality prediction results and reference quality parameters, the quality prediction model is trained, including: The quality prediction model is trained based on the quality prediction results, reference quality parameters and noise influencing factors.

[0103] Noise estimation refers to identifying and quantifying the random errors or uncertainty components in the training sample set. These noises may come from the accuracy limitations of the measurement equipment, changes in the production environment, errors in data recording, etc. Through noise estimation, we can more accurately understand the relationship between real data and random fluctuations in actual data, thereby improving the robustness and generalization ability of the model, avoiding overfitting of the model to the noise data, and thus improving the prediction accuracy.

[0104] The noise estimation method can select different types according to different target production scenarios and different production stages, for example, through statistical analysis, residual analysis and other methods, and the embodiments of this specification do not make specific limitations on this.

[0105] The noise impact factor refers to a parameter obtained after noise estimation that can quantify the degree of influence of noise on the prediction results. The noise impact factor can help the quality prediction model better understand and deal with the uncertainty in the data, ensuring that the prediction results are more reliable.

[0106] After constructing a quality prediction model based on the training sample set, the quality prediction model can be used to estimate the noise of sample quality parameters and sample measurement data to obtain the noise impact factor; after obtaining the noise impact factor, the model can be trained in combination with the noise impact factor, quality prediction results and reference quality parameters.

[0107] By considering the noise impact factor during the quality prediction model training process, the quality prediction model can not only learn based on the direct relationship between input and output data, but also understand how noise affects this relationship, so that the quality prediction model can make more reasonable predictions when faced with new, noisy data. At the same time, during the training process, the loss function can also be adjusted according to the noise impact factor, so that the quality prediction model pays more attention to those data points that are less affected by noise, thereby reducing the interference of noise on the training process.

[0108] For example, the measurement noise of the quality prediction model may be estimated as shown in formula (14): (14) Correspondingly, based on the quality prediction results, the reference quality parameters and the noise influencing factors, the quality prediction model is trained, and the parameters of the model can be adjusted. The specific adjustment method can be shown in formula (15)-formula (16): (15) (16) in, express The covariance matrix at each moment reflects the uncertainty of the model parameters; express The mean vector at time instant reflects the estimated values ​​of the model parameters; is the reference quality parameter; Measure data for the sample; is the noise impact factor. Formula (15) is The covariance matrix at the moment is subtracted from the quality prediction result The related items of update the covariance matrix, reflecting the uncertainty of the updated model parameters; Formula (16) is The mean vector at each moment is subtracted and the quality prediction result is , and combined with the noise impact factor The mean vector is updated based on the relevant terms of , reflecting the updated estimated values ​​of the model parameters.

[0109] After adjusting the parameters of the quality prediction model, the updated model parameters can be used to overwrite the original model parameters to complete the training of the model, and the quality prediction results can be output as the results of the quality prediction model.

[0110] In the embodiments of the present specification, after constructing the quality prediction model, the noise of the sample quality parameters and sample measurement data in the training sample set is estimated to obtain the noise influence factor that quantifies the degree of influence of the noise on the prediction result, and then the quality prediction model is optimized in combination with the noise influence factor, the quality prediction result and the reference quality parameter, so that the quality prediction model can more accurately understand the relationship between the real signal and random fluctuations in the actual data, avoid overfitting of the noise data, improve the prediction accuracy, and further reduce and enhance the stability and reliability of the quality prediction model.

[0111] See also Figure 2 , Figure 2 A flowchart of a product quality prediction method provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0112] Step 202: Acquire current measurement data of key points in the target production scene.

[0113] The current measurement data is the measurement data of key points in the target scene obtained during the product quality prediction process.

[0114] Step 204: input the current measurement data into the quality prediction model to obtain the quality prediction result of the product in the target production scenario, wherein the quality prediction model is trained according to the methods in the above embodiments.

[0115] After obtaining the current measurement data, the current measurement data can be input into the trained quality prediction model to obtain the quality prediction result of the product in the target production scenario. Specifically, the quality prediction model is trained according to the methods in the above embodiments. The specific training process has been described in the above embodiments and will not be repeated here.

[0116] Since the quality prediction model is trained, the future quality prediction results can be accurately obtained based on the current measurement data currently obtained.

[0117] In the embodiments of this specification, by acquiring the current measurement data of key points in the target production scenario and inputting it into the trained quality prediction model, online prediction of product quality is achieved, the accuracy and reliability of the prediction results are ensured, efficient real-time quality monitoring is achieved, the production process is optimized, production costs are reduced, and the consistency and reliability of product quality are ensured.

[0118] In an optional embodiment of the present specification, after inputting the current measurement data into the quality prediction model to obtain the quality prediction result of the product in the target production scenario, it also includes: Obtain reference quality parameters of the corresponding time information of the quality prediction result; Based on the quality prediction result and the reference quality parameter, the model parameters of the quality prediction model are adjusted.

[0119] In the actual production process, due to the influence of various factors such as the production environment, equipment status, and raw material composition, the quality prediction model may also deviate from the actual situation. Therefore, after determining the quality prediction result of a certain time information in the future based on the current measurement data, it is also possible to obtain the reference quality parameters corresponding to the time information for comparative verification to determine whether the output result of the quality prediction model is offset, and based on the quality prediction results and the reference quality parameters, further dynamically adjust the model parameters of the quality prediction model so that the model can be continuously optimized and improved.

[0120] For example, taking the clinker burning scenario in the cement production process as an example, the quality prediction model predicts the free calcium content of a batch of cement clinker at a certain moment in the future based on the currently measured kiln head fire temperature. In order to verify whether this prediction is accurate, the corresponding cement clinker product sample can be obtained at that moment, and laboratory analysis can be used to obtain the actual free calcium content as a reference quality parameter. By comparing the predicted free calcium content with the actual free calcium content, the parameters of the quality prediction model are adjusted to make the quality prediction model more in line with the actual production process.

[0121] In the embodiments of this specification, by obtaining the reference quality parameters of the corresponding time information after the quality prediction results, and dynamically adjusting the model parameters of the quality prediction model based on the comparison between the predicted value and the actual value, the quality prediction model is continuously optimized and improved, the prediction accuracy and stability of the quality prediction model are improved, and the adaptability of the quality prediction model to complex and changeable production environments is enhanced, thereby optimizing the production process, improving overall production efficiency, reducing production costs, and ensuring the consistency and reliability of product quality.

[0122] See also Figure 3 , Figure 3 A flowchart of a product quality prediction method based on a cement production scenario provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0123] Step 302: Acquire current measurement data of key points in the cement production scene.

[0124] Step 304: input the current measurement data into a quality prediction model to obtain a quality prediction result for the free calcium content of cement clinker, wherein the quality prediction model is trained according to the methods in the above embodiments.

[0125] In the cement production scenario, the current measurement data of each key point is obtained. The key points may include parameters such as the kiln head fire temperature, smoke chamber NOx concentration, secondary air temperature, kiln current, etc. The current measurement data may be the measurement data obtained immediately or the measurement data obtained at a certain time.

[0126] The current measurement data is input into the trained quality prediction model to obtain the quality prediction result for the free calcium content of cement clinker. Specifically, the quality prediction model is trained according to the methods in the above embodiments, and the specific training process has been described in the above embodiments and will not be repeated here.

[0127] In the embodiments of this specification, by acquiring the current measurement data of key points such as the kiln head fire temperature, smoke chamber NOx concentration, secondary air temperature, kiln current, etc. in the cement production scene, and inputting it into the trained quality prediction model, online prediction of cement quality is achieved, the accuracy and reliability of the prediction results are ensured, efficient real-time quality monitoring is achieved, the production process is optimized, production costs are reduced, and the consistency and reliability of cement product quality are ensured.

[0128] The following combination Figure 4a-4g To Attachment Figure 5 Taking the application of the quality prediction model training method provided in this specification in the cement production scenario as an example, the quality prediction model training method is further explained. Figure 4a The overall structural diagram and data diagrams of a product quality prediction method provided by an embodiment of this specification are shown. Figure 5 A processing flow chart of a quality prediction model training method provided by an embodiment of this specification is shown.

[0129] like Figure 4a As shown: The key points in the cement production scene, the kiln head fire temperature, the historical measurement data of the kiln current and the actual free calcium content obtained by testing in the cement production scene, among which the kiln head fire temperature fragment is as follows Figure 4b (The horizontal axis x is the time period of free calcium testing, and the vertical axis y is the fire temperature at the kiln head). The kiln current segment is as follows: Figure 4c (x is the time period of free calcium testing, y is the kiln current); according to the sampling time of cement samples corresponding to the actual free calcium content test value, the distribution information of the kiln head fire temperature is estimated through time lag analysis like Figure 4d (The horizontal axis x is the kiln head fire temperature range, and the vertical axis y is the probability density) shows the distribution information of the kiln current like Figure 4e (the horizontal axis x is the kiln current temperature range, and the vertical axis y is its probability density); the kiln head fire temperature distribution information and kiln current distribution information are input into the free calcium prediction model The distribution information of the predicted free calcium content is obtained like Figure 4f As shown (the horizontal axis is the percentage range of free calcium content, and the vertical axis is its probability density); according to the predicted free calcium content and the actual free calcium content, the free calcium content prediction model is trained, that is, the model parameters are updated, and the Bayesian formula can be used specifically Update, the updated model parameter distribution information is as follows Figure 4g(horizontal x and y coordinates are model parameters, and z coordinate is probability density distribution); according to the updated model parameter distribution information, the parameters of the free calcium content prediction model are overwritten to complete the training of the free calcium content prediction model.

[0130] like Figure 5 As shown, the quality prediction model training method specifically includes the following steps.

[0131] Step 502: Obtain historical measurement data of key points in the cement production scenario and reference free calcium content of cement products in the cement production scenario, wherein the key points are points that affect the quality of cement products, and the reference free calcium content is obtained by quality testing of cement products.

[0132] Step 504: Based on the sampling time of the reference free calcium content, determine the target time information of the cement product quality that affects the sampling time through time lag analysis.

[0133] Step 506: According to the target time information, determine the target measurement data of each key point corresponding to the target time information from the historical measurement data of each key point.

[0134] Step 508: Analyze the target measurement data according to the preset time distribution characteristics to obtain distribution information of the target measurement data at each key point.

[0135] Step 510: merging the distribution information of the target measurement data of each key point to obtain the joint distribution information of the target measurement data of each key point.

[0136] Step 512: input the joint distribution information of the target measurement data of each key point into the free calcium content prediction model to obtain the predicted free calcium content of the cement product, wherein the free calcium content prediction model is constructed according to the target sample set obtained by random sampling of the training sample set and obtained by regression processing, and the training sample set includes the sample free calcium content and the sample measurement data.

[0137] Step 514: Based on the predicted free calcium content and the reference free calcium content, a free calcium content prediction model is trained.

[0138] In the embodiments of the present specification, by performing time-lag analysis on historical measurement data and reference free calcium content of key points in the cement production scenario, the target time information affecting product quality is determined, and based on this, the target measurement data and distribution information of each key point are obtained, and further joint distribution information is obtained by merging these data, and inputting them into a free calcium content prediction model trained by random sampling and regression processing, thereby achieving efficient prediction of free calcium content, and dynamically adjusting model parameters by comparing predicted values ​​with actual reference values ​​to ensure continuous optimization and improvement of the model, thereby improving the prediction accuracy and stability of the model, enhancing its adaptability to complex and changeable production environments, and providing solid technical support for intelligent quality control of cement production.

[0139] Corresponding to the above method embodiment, this specification also provides a quality prediction model training device embodiment, Figure 6 FIG. 1 is a schematic diagram showing the structure of a quality prediction model training device provided by an embodiment of the present specification. Figure 6 As shown, the device comprises: The first acquisition module 602 is configured to acquire historical measurement data of key points in the target production scenario and reference quality parameters of products in the target production scenario, wherein the key points are points that affect product quality, and the reference quality parameters are obtained by quality testing of the products; An estimation module 604 is configured to estimate distribution information of target measurement data from historical measurement data through time lag analysis based on a sampling time of a reference quality parameter, wherein the target measurement data is measurement data that affects the reference quality parameter; The first prediction module 606 is configured to input the distribution information of the target measurement data into the quality prediction model to obtain the quality prediction result of the product; The training module 608 is configured to train the quality prediction model based on the quality prediction result and the reference quality parameter.

[0140] Optionally, the estimation module 604 is further configured to: determine the target time information of the product quality that affects the sampling time through time lag analysis based on the sampling time of the reference quality parameter; determine the target measurement data corresponding to the target time information from the historical measurement data according to the target time information; and analyze the target measurement data according to preset time distribution characteristics to obtain distribution information of the target measurement data.

[0141] Optionally, the target production scenario includes at least two key points, and accordingly, the estimation module 604 is further configured to: determine the target measurement data of each key point corresponding to the target time information from the historical measurement data according to the target time information; analyze the target measurement data of each key point respectively according to the preset time distribution characteristics to obtain the distribution information of the target measurement data of each key point; merge the distribution information of the target measurement data of each key point to obtain the distribution information of the target measurement data.

[0142] Optionally, the first prediction module 606 is further configured to: input the distribution information of the target measurement data into the quality prediction model to obtain the predicted quality distribution corresponding to the sampling time; based on the predicted quality distribution and the reference quality parameters, calculate the uncertainty correction amount and the estimated value correction amount of the model parameters of the quality prediction model; and update the model parameters of the quality prediction model according to the uncertainty correction amount and the estimated value correction amount.

[0143] Optionally, the device also includes a construction module configured to: obtain a training sample set, wherein the training sample set includes sample quality parameters and sample measurement data, and the sample measurement data is obtained based on time series matching with the sample quality parameters; and construct a quality prediction model using the training sample set.

[0144] Optionally, the construction module is further configured to: based on the training sample set, randomly sample sample quality parameters and sample measurement data to obtain a target sample set after sampling; use the target sample set to regress the initial quality prediction model to determine the model parameters of the quality prediction model; and construct the quality prediction model based on the model parameters.

[0145] Optionally, the device also includes an impact module configured to: use the quality prediction model to estimate the noise of sample quality parameters and sample measurement data in the training sample set to obtain a noise impact factor; correspondingly, the training module 608 is further configured to: train the quality prediction model based on the quality prediction results, reference quality parameters and noise impact factors.

[0146] The above is a schematic scheme of a quality prediction model training device of this embodiment. It should be noted that the technical scheme of the quality prediction model training device and the technical scheme of the quality prediction model training method described above belong to the same concept, and the details not described in detail in the technical scheme of the quality prediction model training device can be found in the description of the technical scheme of the quality prediction model training method described above.

[0147] Corresponding to the above method embodiment, this specification also provides a product quality prediction device embodiment, Figure 7 FIG. 1 is a schematic diagram showing the structure of a product quality prediction device provided by an embodiment of the present specification. Figure 7 As shown, the device comprises: The second acquisition module 702 is configured to acquire current measurement data of key points in the target production scene.

[0148] The second prediction module 704 is configured to input the current measurement data into a quality prediction model to obtain a quality prediction result of the product in a target production scenario, wherein the quality prediction model is trained according to the methods in the above embodiments.

[0149] Optionally, the device further includes an adjustment module configured to: obtain reference quality parameters of time information corresponding to the quality prediction result; and adjust model parameters of the quality prediction model based on the quality prediction result and the reference quality parameters.

[0150] The above is a schematic scheme of a product quality prediction device of this embodiment. It should be noted that the technical scheme of the product quality prediction device and the technical scheme of the above-mentioned product quality prediction method belong to the same concept, and the details not described in detail in the technical scheme of the product quality prediction device can be referred to the description of the technical scheme of the above-mentioned product quality prediction method.

[0151] Corresponding to the above method embodiment, this specification also provides an embodiment of a product quality prediction device based on a cement production scenario. Figure 8 FIG. 1 is a schematic diagram showing a structural diagram of a product quality prediction device based on a cement production scenario provided by an embodiment of this specification. Figure 8 As shown, the device comprises: The third acquisition module 802 is configured to acquire current measurement data of key points in the cement production scene; The third prediction module 804 is configured to input the current measurement data into a quality prediction model to obtain a quality prediction result for the free calcium content of cement clinker, wherein the quality prediction model is trained according to the methods in the above embodiments.

[0152] The above is a schematic scheme of a product quality prediction device based on a cement production scenario in this embodiment. It should be noted that the technical scheme of the product quality prediction device based on the cement production scenario and the technical scheme of the product quality prediction method based on the cement production scenario belong to the same concept. For details not described in detail in the technical scheme of the product quality prediction device based on the cement production scenario, please refer to the description of the technical scheme of the product quality prediction method based on the cement production scenario.

[0153] Fig. 9The structure block diagram of a computing device 900 provided according to an embodiment of the present specification is shown. The components of the computing device 900 include but are not limited to a memory 910 and a processor 920. The processor 920 is connected to the memory 910 via a bus 930, and a database 950 is used to store data.

[0154] The computing device 900 also includes an access device 940 that enables the computing device 900 to communicate via one or more networks 960. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 940 may include one or more of any type of network interface (e.g., a network interface card (NIC)) that is wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a world-wide interoperability for microwave access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, and a near field communication (NFC).

[0155] In one embodiment of the present specification, the above components of the computing device 900 and Fig. 9 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Fig. 9 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0156] The computing device 900 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 900 may also be a mobile or stationary server.

[0157] The processor 920 is used to execute the following computer program / instruction, which implements the steps of the above methods when executed by the processor. The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the above methods belong to the same concept, and the details of the technical scheme of the computing device that are not described in detail can be found in the description of the technical scheme of the above methods.

[0158] An embodiment of this specification also provides a computer-readable storage medium, which stores a computer program / instruction, and the steps of the above-mentioned methods are implemented when the computer program / instruction is executed by the processor. The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical solution of the storage medium belongs to the same concept as the technical solution of the above-mentioned methods. For the details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the above-mentioned methods.

[0159] This specification also provides a computer program product in one embodiment, including a computer program / instruction, which implements the steps of the above methods when executed by a processor. The above is a schematic scheme of a computer program product of this embodiment. It should be noted that the technical solution of the computer program product and the technical solution of the above methods belong to the same concept, and the details of the technical solution of the computer program that are not described in detail can be found in the description of the technical solution of the above methods.

[0160] The above is a description of a specific embodiment of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0161] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electric carrier signals and telecommunication signals. It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited to the described order of actions, because according to the embodiments of this specification, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0162] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can understand and use this specification well. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A quality prediction model training method, comprising: Acquire historical measurement data of key points in a target production scenario and reference quality parameters of products in the target production scenario, wherein the key points are points that affect product quality, and the reference quality parameters are obtained by quality testing of the products; Based on the sampling time of the reference quality parameter, the distribution information of the target measurement data is estimated from the historical measurement data by time lag analysis, wherein the target measurement data is the measurement data that affects the reference quality parameter; Inputting the distribution information of the target measurement data into a quality prediction model to obtain a quality prediction result of the product; The quality prediction model is trained based on the quality prediction result and the reference quality parameter.

2. The method according to claim 1, wherein the step of estimating the distribution information of the target measurement data from the historical measurement data based on the sampling time of the reference quality parameter by time lag analysis comprises: Based on the sampling time of the reference quality parameter, determining target time information of product quality affecting the sampling time through time lag analysis; According to the target time information, determining the target measurement data corresponding to the target time information from the historical measurement data; The target measurement data is analyzed according to preset time distribution characteristics to obtain distribution information of the target measurement data.

3. According to the method of claim 2, the target production scene includes at least two key points; and determining the target measurement data corresponding to the target time information from the historical measurement data according to the target time information comprises: According to the target time information, determining the target measurement data of each key point corresponding to the target time information from the historical measurement data; The step of analyzing the target measurement data according to the preset time distribution characteristics to obtain distribution information of the target measurement data includes: According to the preset time distribution characteristics, the target measurement data of each key point is analyzed respectively to obtain the distribution information of the target measurement data of each key point; The distribution information of the target measurement data of each key point is merged to obtain the distribution information of the target measurement data.

4. According to the method according to any one of claims 1 to 3, inputting the distribution information of the target measurement data into a quality prediction model to obtain the quality prediction result of the product comprises: Inputting the distribution information of the target measurement data into a quality prediction model to obtain a predicted quality distribution corresponding to the sampling time; The training of the quality prediction model based on the quality prediction result and the reference quality parameter includes: Based on the predicted quality distribution and the reference quality parameter, an uncertainty correction amount and an estimated value correction amount of a model parameter of the quality prediction model are calculated; The model parameters of the quality prediction model are updated according to the uncertainty correction amount and the estimated value correction amount.

5. The method according to claim 1, before inputting the distribution information of the target measurement data into the quality prediction model to obtain the quality prediction result of the product, further comprises: Acquire a training sample set, wherein the training sample set includes a sample quality parameter and sample measurement data, and the sample measurement data is obtained based on a time series matching with the sample quality parameter; A quality prediction model is constructed using the training sample set.

6. The method according to claim 5, wherein the step of constructing a quality prediction model using the training sample set comprises: Based on the training sample set, randomly sampling the sample quality parameter and the sample measurement data to obtain a sampled target sample set; Using the target sample set, performing regression processing on the initial quality prediction model to determine model parameters of the quality prediction model; Based on the model parameters, the quality prediction model is constructed.

7. The method according to claim 5 or 6, after using the training sample set to construct the quality prediction model, further comprising: Using the quality prediction model, performing noise estimation on sample quality parameters and sample measurement data in the training sample set to obtain a noise impact factor; The training of the quality prediction model based on the quality prediction result and the reference quality parameter includes: The quality prediction model is trained based on the quality prediction result, the reference quality parameter and the noise impact factor.

8. A product quality prediction method comprising: Obtain current measurement data of key points in the target production scenario; The current measurement data is input into a quality prediction model to obtain a quality prediction result of the product in the target production scenario, wherein the quality prediction model is trained according to the method described in any one of claims 1-7.

9. The method according to claim 8, after inputting the current measurement data into the quality prediction model to obtain the quality prediction result of the product in the target production scenario, further comprising: Obtaining a reference quality parameter of the time information corresponding to the quality prediction result; Based on the quality prediction result and the reference quality parameter, a model parameter of the quality prediction model is adjusted.

10. A product quality prediction method based on cement production scenario, comprising: Obtain current measurement data of key points in cement production scenarios; The current measurement data is input into a quality prediction model to obtain a quality prediction result for the free calcium content of cement clinker, wherein the quality prediction model is trained according to the method according to any one of claims 1-7.

11. A quality prediction model training device, comprising: A first acquisition module is configured to acquire historical measurement data of key points in a target production scenario and reference quality parameters of products in the target production scenario, wherein the key points are points that affect product quality, and the reference quality parameters are obtained by performing quality inspection on the products; an estimation module, configured to estimate distribution information of target measurement data from the historical measurement data through time lag analysis based on a sampling time of the reference quality parameter, wherein the target measurement data is measurement data that affects the reference quality parameter; A first prediction module is configured to input the distribution information of the target measurement data into a quality prediction model to obtain a quality prediction result of the product; The training module is configured to train the quality prediction model based on the quality prediction result and the reference quality parameter.

12. A product quality prediction device, comprising: A second acquisition module is configured to acquire current measurement data of key points in a target production scene; The second prediction module is configured to input the current measurement data into a quality prediction model to obtain a quality prediction result of the product under the target production scenario, wherein the quality prediction model is trained according to the method described in any one of claims 1-7.

13. A product quality prediction device based on cement production scenarios, comprising: A third acquisition module is configured to acquire current measurement data of key points in the cement production scene; The third prediction module is configured to input the current measurement data into a quality prediction model to obtain a quality prediction result for the free calcium content of cement clinker, wherein the quality prediction model is trained according to the method described in any one of claims 1-7.

14. A computing device comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the method described in any one of claims 1 to 10 are implemented.

15. A computer-readable storage medium storing a computer program / instruction, wherein the computer program / instruction, when executed by a processor, implements the steps of the method according to any one of claims 1 to 10.

16. A computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 10.

Images