CN114925934A - Prediction model verification method for building ceramic production - Google Patents

Abstract

The invention relates to the technical field of model verification, in particular to a prediction model verification method for architectural ceramic production, which comprises the following steps: step S1: performing off-line verification on the prediction model to be verified to obtain an off-line verification result; step S2: judging whether the offline verification result meets the offline production requirement, if so, performing step S3, if not, adjusting the prediction model, and repeating step S1; step S3: performing online verification on the prediction model passing the offline verification to obtain an online verification result; step S4: and judging whether the online verification result meets the online production requirement, if so, finishing the verification, if not, adjusting the prediction model, and repeating the step S1. The method can verify the feasibility and the accuracy of the prediction model for building ceramic production, guarantee the effectiveness of the prediction model in the verification process and reduce the trial and error cost.

Description

A Predictive Model Verification Method for Architectural Ceramics Production

technical field

The invention relates to the technical field of model verification, in particular to a prediction model verification method for building ceramic production.

Background technique

With the continuous improvement of my country's national economic level and the continuous progress of social productivity, more and more ceramic furniture is used for home decoration. When people choose many different brands and manufacturers, they are more and more concerned about the variety and quality of ceramic furniture. The more concerned. As a product that must be used for product decoration, ceramic tiles are of varying quality. How to accurately detect the quality of ceramic tiles has become an important issue in construction engineering.

At present, in order to achieve the evaluation of the quality of ceramic tiles, and to better improve the quality analysis of ceramic tiles, the building ceramics big data prediction model came into being. Model analysis of the production state and the adjustment of predicted production parameters to achieve advance estimation of ceramic quality indicators, but this kind of prediction model relies on the rigid generation of existing data and is not suitable for actual production. A method is needed to verify its feasibility and performance. accuracy.

SUMMARY OF THE INVENTION

The purpose of the present invention is to propose a predictive model verification method for the production of architectural ceramics, which can verify the feasibility and accuracy of the predictive model for the production of architectural ceramics, while ensuring the validity of the predictive model in the verification process and reducing the cost of trial and error.

For this purpose, the present invention adopts the following technical solutions:

A predictive model validation method for architectural ceramics production, comprising the following steps:

Step S1: perform offline verification on the prediction model to be verified, and obtain an offline verification result;

Step S2: judging whether the offline verification result meets the offline production requirements, if so, proceed to step S3, if not, adjust the prediction model, and repeat step S1;

Step S3: perform online verification on the prediction model that has passed the offline verification, and obtain an online verification result;

Step S4: Determine whether the online verification result meets the online production requirements, if yes, complete the verification, if not, adjust the prediction model, and repeat step S1.

Preferably, in step S1, the prediction model to be verified is subjected to offline verification to obtain an offline verification result, including the following steps:

Step S11: Input the N actual production historical data into the prediction model to be verified, obtain N offline predicted production parameters, and draw the N predicted production parameters into an offline predicted production curve;

Step S12: Drawing the actual production history parameters corresponding to the N actual production history data into an actual production history curve;

Step S13: Draw the upper limit curve of predicted production and the lower limit curve of predicted production according to the upper and lower limits of the prediction standard of the prediction model, and define the upper limit curve of predicted production and the lower limit curve of predicted production; wherein the area between the upper limit curve of predicted production and the lower limit curve of predicted production is defined as the prediction area ;

Step S14: drawing the upper limit curve of the process standard and the lower limit curve of the process standard according to the upper and lower limits of the production process standard, wherein the area between the upper limit curve of the process standard and the lower limit curve of the process standard is defined as a qualified area;

Step S15: Draw the offline predicted production curve, the actual production history curve, the predicted production upper limit curve, the predicted production lower limit curve, the process standard upper limit curve and the process standard lower limit curve together into a graph to obtain an offline verification result.

Preferably, in step S2, the judging whether the offline verification result meets the offline production requirements includes the following steps:

Step S21: judging whether the actual production history curve is within the prediction area;

Step S22: Determine whether the upper limit curve of predicted production and the lower limit curve of predicted production are within the qualified area.

Preferably, in step S21, the judging whether the actual production history curve is within the prediction area includes the following steps:

Step S211: Calculate the predicted proportion that the actual production history curve is not within the predicted area;

Step S212: Determine whether the prediction ratio is smaller than the prediction error determination point, if the prediction ratio is smaller than the prediction error determination point, then execute step S22; if the prediction ratio is greater than or equal to the prediction error determination point, output the error of the prediction model. information, modify the prediction model, and repeat step S1.

Preferably, in step S22, the judging whether the predicted production upper limit curve and the predicted production lower limit curve are within the qualified area includes the following steps:

Step S221: Calculate the qualified proportion that is not within the qualified area;

Step S222: determine whether the qualified proportion is less than the qualified error judgment point, if the qualified proportion is less than the qualified error judgment point, output the prediction model through offline verification information, and execute step S3; if the qualified proportion is greater than or equal to the qualified error judgment point, then Output the information that the prediction model is wrong, and repeat step S1.

Preferably, in step S3, online verification is performed on the prediction model that has passed the offline verification, and an online verification result is obtained, including the following steps:

Step S31: Input the N real-time production data into the offline-verified prediction model to obtain N online predicted production parameters;

Step S32: Drawing the N online predicted production parameters into an online production curve to obtain an online verification result.

Preferably, in step S4, the judging whether the online verification result meets the online production requirements includes the following steps:

Step S41: send the online verification result to the expert module, and receive expert judgment information;

Step S42: If the expert judges that the information is in line, the production is carried out according to the online predicted production parameters, the actual real parameters during production are recorded, and steps S43-S44 are executed;

If the expert judges that the information is inconsistent, adjust the prediction model, and repeat step S1;

Step S43: Send the actual real parameters and the online predicted production parameters to the expert module to obtain difference analysis information;

Step S44: Optimize the prediction model according to the difference analysis result, and repeat step S3 until the difference between the actual real parameter and the online predicted production parameter is within the set allowable range, then the prediction model is judged to be valid.

Preferably, the difference analysis information includes offline test difference analysis information and online test difference analysis information;

The off-line test difference analysis information includes the parameter error rate between the actual real parameter and the online predicted production parameter, and the parameter error rate is (online predicted production parameter-actual real parameter)/actual real parameter*100%;

The online test difference analysis information includes a production efficiency difference result and a quality difference result;

The production efficiency difference result is the production efficiency difference result between the actual production efficiency obtained from the actual real parameters and the predicted production efficiency obtained from the online predicted production parameters, and the production efficiency difference result is (predicted production efficiency-actual production efficiency)/ Actual production efficiency*100%;

The quality difference result is the quality difference result between the actual quality obtained by the actual real parameters and the predicted quality obtained by predicting the production parameters online, and the quality difference result is (predicted quality-actual quality)/actual quality*100%.

One of the above technical solutions has the following beneficial effects: first, after offline verification meets the offline production requirements, the prediction model is verified online based on the established online production requirements; To meet the process production requirements and forecast production requirements designed in the production requirements, not only to meet the verification operability and rationality of the forecast model, but also to predict that the forecast data will gradually meet the production requirements, so as to better evaluate the quality of the actual ceramic tiles produced. Evaluation and Analysis.

Description of drawings

Fig. 1 is the flow chart of the predictive model verification method that the present invention is used for building ceramics production;

Fig. 2 is the principle schematic diagram of the predictive model verification method that the present invention is used for building ceramics production;

3 is a schematic diagram of the offline verification result of Embodiment 1 of the predictive model verification method for building ceramic production according to the present invention;

4 is a schematic diagram of the offline verification result of Embodiment 2 of the predictive model verification method for building ceramic production according to the present invention;

5 is a schematic diagram of the online verification result of Embodiment 3 of the predictive model verification method for building ceramic production according to the present invention;

6 is a schematic diagram of the online verification result of Embodiment 4 of the predictive model verification method for building ceramic production according to the present invention;

Detailed ways

The technical solutions of the present invention are further described below with reference to the accompanying drawings and through specific embodiments.

As shown in Figure 1, a predictive model validation method for architectural ceramics production includes the following steps:

Step S1: perform offline verification on the prediction model to be verified, and obtain an offline verification result;

Step S2: judging whether the offline verification result meets the offline production requirements, if so, proceed to step S3, if not, adjust the prediction model, and repeat step S1;

Step S3: perform online verification on the prediction model that has passed the offline verification, and obtain an online verification result;

Step S4: Determine whether the online verification result meets the online production requirements, if yes, complete the verification, if not, adjust the prediction model, and repeat step S1.

How to verify the feasibility of the prediction model used for the production of architectural ceramics, and at the same time to ensure the validity of the model in the verification process and reduce the cost of trial and error, the verification method of the prediction model in this embodiment is to first verify that it meets the offline production requirements, and then Based on the established online production requirements, the prediction model is verified online; through multiple adjustments and judging whether the prediction data of the prediction model meets the process production requirements and predicted production requirements designed in the production requirements, it not only satisfies the verification operability of the prediction model At the same time, it is predicted that the forecast data will gradually meet the requirements of production, so as to better evaluate and analyze the quality of the actual ceramic tiles produced.

To further illustrate, in step S1, the prediction model to be verified is subjected to offline verification to obtain an offline verification result, including the following steps:

Step S11: Input the N actual production historical data into the prediction model to be verified, obtain N offline predicted production parameters, and draw the N predicted production parameters into an offline predicted production curve;

Step S12: Drawing the actual production history parameters corresponding to the N actual production history data into an actual production history curve;

Step S13: Draw the upper limit curve of predicted production and the lower limit curve of predicted production according to the upper and lower limits of the prediction standard of the prediction model, and define the upper limit curve of predicted production and the lower limit curve of predicted production; wherein the area between the upper limit curve of predicted production and the lower limit curve of predicted production is defined as the prediction area ;

Step S14: drawing the upper limit curve of the process standard and the lower limit curve of the process standard according to the upper and lower limits of the production process standard, wherein the area between the upper limit curve of the process standard and the lower limit curve of the process standard is defined as a qualified area;

Step S15: Draw the offline predicted production curve, the actual production history curve, the predicted production upper limit curve, the predicted production lower limit curve, the process standard upper limit curve and the process standard lower limit curve together into a graph to obtain an offline verification result.

As shown in FIG. 2 , in this embodiment, the offline verification result is drawn on a graph by using a plurality of parameter data, and the changes of the statistical data can be fed back intuitively.

Further explanation, in step S2, the described judgment whether the offline verification result meets the offline production requirements, includes the following steps:

Step S21: judging whether the actual production history curve is within the prediction area;

Step S22: Determine whether the upper limit curve of predicted production and the lower limit curve of predicted production are within the qualified area.

In this embodiment, the verification quality of the offline verification is improved through continuous judgment in the offline verification.

To further illustrate, in step S21, the judging whether the actual production history curve is within the prediction area includes the following steps:

Step S211: Calculate the predicted proportion that the actual production history curve is not within the predicted area;

Step S212: Determine whether the prediction ratio is smaller than the prediction error determination point, if the prediction ratio is smaller than the prediction error determination point, then execute step S22; if the prediction ratio is greater than or equal to the prediction error determination point, output the error of the prediction model. information, modify the prediction model, and repeat step S1.

Further explanation, in step S22, the described judgment of whether the predicted production upper limit curve and the predicted production lower limit curve are within the qualified area includes the following steps:

Step S221: Calculate the qualified proportion that is not within the qualified area;

Step S222: determine whether the qualified proportion is less than the qualified error judgment point, if the qualified proportion is less than the qualified error judgment point, output the prediction model through offline verification information, and execute step S3; if the qualified proportion is greater than or equal to the qualified error judgment point, then Output the information that the prediction model is wrong, and repeat step S1.

In this embodiment, the offline predicted production parameters are compared and analyzed with the production process standards and actual production historical parameters, so as to further improve the verification quality of the offline verification. The process includes drawing the N offline predicted production parameters output by the prediction model into the upper limit curve of predicted production and the lower limit curve of predicted production, and it is judged that all of them fall within the upper limit curve of the process standard and the lower limit curve of the process standard required by the production process standard, and are consistent with the curve of the lower limit of the process standard. There is no statistical difference between the actual production history curves of the actual production history parameters, that is, 95% of the actual production history parameters fall within the predicted parameter range output by the model, indicating that the model is available and the worker experience can be learned. , so that the offline verification results meet the actual production requirements.

Further description, in step S3, online verification is performed on the prediction model that has passed the offline verification, and the online verification result is obtained, including the following steps:

Step S31: Input the N real-time production data into the offline-verified prediction model to obtain N online predicted production parameters;

Step S32: Drawing the N online predicted production parameters into an online production curve to obtain an online verification result.

Further explanation, in step S4, described judging whether the online verification result meets the online production requirements, including the following steps:

Step S41: send the online verification result to the expert module, and receive expert judgment information;

Step S42: If the expert judges that the information is in line, the production is carried out according to the online predicted production parameters, the actual real parameters during production are recorded, and steps S43-S44 are executed;

If the expert judges that the information is inconsistent, adjust the prediction model, and repeat step S1;

Step S43: Send the actual real parameters and the online predicted production parameters to the expert module to obtain difference analysis information;

Step S44: Optimize the prediction model according to the difference analysis result, and repeat step S3 until the difference between the actual real parameter and the online predicted production parameter is within the set allowable range, then the prediction model is judged to be valid.

Since using the model to predict parameters may lead to unstable production or even production quality accidents, it is necessary to pass multiple offline verifications before online verification to meet production requirements, and at the same time to ensure the same production conditions. Secondly, the online predicted production parameters predicted by the prediction model need to be discussed and verified with process experts and sintering experts. If it conforms to the production process, then adjust and produce according to the online predicted production parameters. At the same time, big data model experts, process experts, firing experts and quality control experts track online to prevent production accidents. At the same time, the production changes, especially the final product quality status, are recorded at all times, and product quality defects unrelated to the kiln are identified to judge the accuracy of the online forecast production parameters predicted by the forecast model.

Preferably, the difference analysis information includes offline test difference analysis information and online test difference analysis information;

The off-line test difference analysis information includes the parameter error rate between the actual real parameter and the online predicted production parameter, and the parameter error rate is (online predicted production parameter-actual real parameter)/actual real parameter*100%;

The online test difference analysis information includes a production efficiency difference result and a quality difference result;

The production efficiency difference result is the production efficiency difference result between the actual production efficiency obtained from the actual real parameters and the predicted production efficiency obtained from the online predicted production parameters, and the production efficiency difference result is (predicted production efficiency-actual production efficiency)/ Actual production efficiency*100%;

The quality difference result is the quality difference result between the actual quality obtained by the actual real parameters and the predicted quality obtained by predicting the production parameters online, and the quality difference result is (predicted quality-actual quality)/actual quality*100%.

To further illustrate the operability and rationality of the verification of the present invention, four embodiments are provided below. Example 1: Offline Verification of Ball Milling Prediction Model

Table 1 Off-line verification results of ball mill prediction model for predicting mud fineness

Summary: The offline verification pass rate of the ball mill model predicting the mud quality is 90%, and its normal production product index is 90%. Therefore, the offline prediction production parameters of the prediction model meet the offline production requirements, indicating that the ball mill prediction model meets the production requirements.

Example 2: Offline Verification of Furnace Prediction Model

Table 2 The offline verification results of the predicted temperature of the furnace prediction model

Summary: The temperature predicted by the kiln model is within the set range of the process standard temperature, and the actual production trend is basically the same, indicating that the kiln model prediction meets the production process requirements.

Example 3: Validation of the Ball Milling Prediction Model

Table 3 Online verification results of ball mill prediction model predicting mud fineness

Summary: According to the above results, it is found that the predicted mud fineness is basically consistent with the actual mud fineness, the absolute value of the error rate is 4% at most, and the error rate is ≤5%, indicating that the model effect meets the requirements, so it shows that the ball milling fineness prediction model conforms to the production process. requirements, and the prediction accuracy is high.

Example 4: Online verification of furnace prediction model

Table 4 The online verification results of the production excellence rate of the prediction parameters of the kiln model

Summary: According to the daily excellence rate error value in the normal production and verification process, it is found that the error value is within the range of ±5, indicating that the kiln prediction model is available.

The technical principle of the present invention has been described above with reference to the specific embodiments. These descriptions are only for explaining the principle of the present invention, and should not be construed as limiting the protection scope of the present invention in any way. Based on the explanations herein, those skilled in the art can think of other specific embodiments of the present invention without creative efforts, and these equivalent modifications or substitutions are all included within the scope defined by the claims of the present application.

Claims (8)

1. A predictive model validation method for architectural ceramic production, comprising the steps of:

step S1: performing off-line verification on the prediction model to be verified to obtain an off-line verification result;

step S2: judging whether the offline verification result meets the offline production requirement, if so, performing step S3, if not, adjusting the prediction model, and repeating step S1;

step S3: performing online verification on the prediction model passing the offline verification to obtain an online verification result;

step S4: and judging whether the online verification result meets the online production requirement, if so, finishing the verification, if not, adjusting the prediction model, and repeating the step S1.

2. The method for verifying the prediction model for the production of architectural ceramics according to claim 1, wherein in step S1, the prediction model to be verified is verified offline to obtain an offline verification result, and the method comprises the following steps:

step S11: inputting N pieces of historical data of actual production into a prediction model to be verified to obtain N pieces of offline prediction production parameters, and drawing the N pieces of prediction production parameters into an offline prediction production curve;

step S12: drawing actual production history parameters corresponding to the historical data of the N actual productions into an actual production history curve;

step S13: drawing a predicted production upper limit curve and a predicted production lower limit curve according to the upper limit and the lower limit of the prediction standard of the prediction model, and predicting the production upper limit curve and the predicted production lower limit curve; wherein a region between the predicted production upper limit curve and the predicted production lower limit curve is defined as a predicted region;

step S14: drawing a process standard upper limit curve and a process standard lower limit curve according to the upper limit and the lower limit of the production process standard, wherein the region between the process standard upper limit curve and the process standard lower limit curve is defined as a qualified region;

step S15: and drawing an offline predicted production curve, an actual production history curve, a predicted production upper limit curve, a predicted production lower limit curve, a process standard upper limit curve and a process standard lower limit curve into a curve graph together to obtain an offline verification result.

3. The method for verifying the prediction model for the production of architectural ceramics according to claim 2, wherein in step S2, the step of determining whether the off-line verification result meets the off-line production requirement comprises the following steps:

step S21: judging whether the actual production historical curve is in the prediction area or not;

step S22: and judging whether the predicted production upper limit curve and the predicted production lower limit curve are in qualified areas.

4. The predictive model verification method for architectural ceramic production according to claim 3, wherein in step S21, said judging whether the actual production history curve is within the predicted area comprises the steps of:

step S211: calculating the prediction ratio of the actual production historical curve not in the prediction region;

step S212: judging whether the prediction ratio is smaller than a prediction error judgment point, and if the prediction ratio is smaller than the prediction error judgment point, executing a step S22; if the prediction percentage is greater than or equal to the prediction error determination point, outputting information that the prediction model is incorrect, modifying the prediction model, and repeating step S1.

5. The predictive model verification method for architectural ceramic production of claim 3, wherein in step S22, the determining whether the predicted production upper limit curve and the predicted production lower limit curve are within the qualified area comprises the steps of:

step S221: calculating the qualified proportion which is not in the qualified area;

step S222: judging whether the qualified proportion is smaller than a qualified error judgment point, if so, outputting offline verification information of the prediction model, and executing the step S3; if the percentage of qualified products is greater than or equal to the qualified error determination point, the information that the prediction model is erroneous is output, and step S1 is repeatedly executed.

6. The method for verifying the prediction model for the production of architectural ceramics according to claim 1, wherein in step S3, the prediction model verified off-line is verified on-line to obtain an on-line verification result, comprising the following steps:

step S31: inputting the N real-time production data into a prediction model passing offline verification to obtain N online prediction production parameters;

step S32: and drawing the N online predicted production parameters into an online production curve to obtain an online verification result.

7. The prediction model verification method for architectural ceramic production according to claim 6, wherein in step S4, the step of judging whether the online verification result meets the online production requirement comprises the following steps:

step S41: sending the online verification result to an expert module, and receiving expert judgment information;

step S42: if the expert judges that the information is in line, producing according to the on-line prediction production parameters, recording actual real parameters during production, and executing the steps S43-S44;

if the expert judges that the information does not conform to the preset information, adjusting the prediction model, and repeatedly executing the step S1;

step S43: sending the actual real parameters and the online predicted production parameters to an expert module to obtain difference analysis information;

step S44: and optimizing the prediction model according to the difference analysis information, and repeatedly executing the step S3 until the difference between the actual real parameter and the online predicted production parameter is within the set allowable range, judging that the prediction model has validity, and completing verification.

8. The predictive model validation method for architectural ceramic production of claim 7, wherein the difference analysis information includes off-line test difference analysis information and on-line test difference analysis information;

the off-line test difference analysis information comprises a parameter error rate of actual real parameters and on-line predicted production parameters, wherein the parameter error rate is (on-line predicted production parameters-actual real parameters)/actual real parameters 100%;

the online test difference analysis information comprises a production efficiency difference result and a quality difference result;

the production efficiency difference result is a production efficiency difference result of actual production efficiency obtained by actual real parameters and predicted production efficiency obtained by online prediction of production parameters, and the production efficiency difference result is (predicted production efficiency-actual production efficiency)/actual production efficiency 100%;

the quality difference result is a quality difference result of an actual quality obtained from the actual real parameters and a predicted quality obtained from the online predicted production parameters, and the quality difference result is (predicted quality-actual quality)/actual quality x 100%.

Images