High-efficiency adaptive complex industrial parameter prediction method based on regression application scoring

Abstract

The invention discloses a high-efficiency adaptive complex industrial parameter prediction method based on regression application scoring, and belongs to the field of industrial parameter prediction. According to the feature screening algorithm based on regression application scoring, the value of each feature is comprehensively evaluated by integrating three key indexes, namely the hit rate with the error within 3%, the decision coefficient R2 and the time consumption score, so that the regression application score is obtained, the actual contribution of each feature to the prediction performance of the model can be accurately evaluated, and the prediction efficiency of the model is improved. Rapid and effective feature selection is realized; and meanwhile, a mode of iteratively updating the model is provided, the data in a period of time is utilized to train the model to predict the data in a period of time with a short distance, and the newly generated data is periodically utilized to retrain the model to predict the data in the later period of time, so that the problem of model timeliness is solved, and the model can stably and accurately predict for a long time.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial parameter prediction, and more specifically, to a high-efficiency adaptive complex industrial parameter prediction method based on regression application scoring. Background Art

[0002] There are many key process indicators that need to be predicted in industrial scenarios. For example, in the steel industry, drum strength, ignition temperature, etc. are important factors affecting the quality of the final product. These key indicators play a vital role in sinter quality, energy consumption, and output. However, many key indicators cannot be directly obtained through online measurement, so forecasting or soft measurement models are needed to indirectly obtain these data. Establishing a mapping relationship between key indicators and process parameters can effectively solve this problem.

[0003] Existing modeling methods mainly include mechanism models based on physical principles, intelligent models based on data-driven, etc. However, due to the complex and changeable industrial production environment, the many process parameters involved, and the influence of factors such as equipment aging, most models are fixed after the initial training and are difficult to adapt to the above changes, which leads to the problem that the prediction accuracy of the model will decrease significantly over time. Summary of the invention

[0004] 1. Technical problem to be solved by the invention

[0005] In order to solve the problem that the industrial parameter measurement model in the prior art will become ineffective after long-term application, the present invention intends to provide a high-efficiency adaptive complex industrial parameter prediction method based on regression application scoring, which can perform accurate prediction in a long-term and stable manner and alleviate computing pressure.

[0006] 2. Technical solution

[0007] In order to solve the problem that traditional measurement models in the industry will become ineffective after long-term application, the present invention innovatively proposes a method of iteratively updating the model, that is, using the data of a period of time to train the model to predict the data in a shorter period of time in the future, and then regularly using the newly generated data to retrain the model to predict the data in the next period of time, rather than using the model for prediction after training a model. This method solves the problem of model timeliness, enabling it to make accurate predictions in a long-term and stable manner, but iteratively updating the training model requires frequent data collection for calculation, which puts a lot of computing pressure on the server and is difficult to adapt to the requirements of real-time control of field equipment. In order to alleviate the computing pressure, the present invention further proposes a feature screening algorithm based on regression application scoring. The algorithm integrates three key indicators - the hit rate within an error of 3%, the determination coefficient R 2And the time-consuming score is used to comprehensively evaluate the value of each feature, and then the regression application score is obtained. With the help of the regression application score as an evaluation indicator, the algorithm can accurately evaluate the actual contribution of each feature to the model prediction performance, and realize fast and effective feature selection. This optimization strategy effectively overcomes the limitations of traditional feature screening algorithms. For example, although the features selected by the Pearson correlation algorithm and the Spearman correlation algorithm may be mathematically related, they may not necessarily make positive contributions in the actual model training process. The algorithm proposed in the present invention can not only eliminate useless or redundant features, reduce dependence on computing resources, and greatly improve the computing speed, but also significantly improve the generalization ability and prediction accuracy of the model, and enhance the application effect of the model.

[0008] Specifically, the present invention provides a high-efficiency adaptive complex industrial parameter prediction method based on regression application scoring, the process is as follows:

[0009] Step 1: Data acquisition and preprocessing

[0010] (1) Data collection: PLC data collection equipment is installed at the equipment location where data needs to be collected in the industry. For example, if the opening of a pipeline pressure valve needs to be collected, the PLC device is installed at the valve in the pipeline. The PLC data collection equipment ensures that data can be collected quickly, accurately and stably.

[0011] (2) Data preprocessing: After receiving the data, the data quality is checked first, especially to detect whether there are missing values. Once missing values ​​are found, the values ​​of the nearest neighboring data points are used to fill them in to maintain the integrity and consistency of the data set. Next, in order to ensure the reliability of the data, the isolation forest technology is used to perform outlier detection on the data; in the process of building the isolation forest model, an important parameter is the number of decision trees in the forest. In the present invention, the number of decision trees is set to 120 to ensure that the model has sufficient complexity and detection accuracy. The isolation forest model works by randomly splitting the data, and each split will generate a path from the root node to the leaf node. Outliers are usually more easily isolated, so data points with shorter path lengths are given lower anomaly scores, which means that they are more likely to be outliers. After completing the anomaly scoring of all data points, a threshold is set according to the scoring results to delete outliers. In industrial practice, the commonly used threshold is set to 0.01, which means that approximately 1 out of every 100 data points will be deleted as an outlier. This process helps to purify the data set and ensure the accuracy and reliability of subsequent model training. The isolation forest algorithm can initially eliminate abnormal data, but this is far from enough. The present invention has found through in-depth practical observations that in an industrial environment, factors such as machine failures may cause data anomalies over a period of time. The isolation forest algorithm can effectively detect abnormal data when normal and abnormal working conditions alternate, but the effect is not ideal when processing continuous abnormal data. Therefore, the present invention further discovered, based on an in-depth analysis of the process flow, that certain key features in industrial processes should not have a value of 0. Based on this standard, a deeper level of cleaning of outliers is performed, and data samples that should not have a value of 0 but have a value of 0 are deleted. This series of measures is intended to improve data quality and ensure the accuracy and reliability of modeling.

[0012] Step 2: Feature Screening

[0013] After completing the data preprocessing, the feature scoring step will be performed next. This invention introduces a regression application scoring method based on three key indicators: the hit rate within the error range of 3%, the determination coefficient R 2 Each indicator is multiplied by its corresponding weight - the hit rate is multiplied by a, R 2 The sum of these weights is 1, that is, a+b+c=1. The specific values ​​of the weight coefficients a, b, and c need to be adjusted and determined according to the actual application scenario. In different application scenarios, the focus of the model will be different. If the prediction accuracy is the most important, the coefficient a of the hit rate will be set higher; on the contrary, if the generalization ability or computational efficiency of the model is more concerned, the coefficient a of the hit rate will be appropriately reduced and the R 2The coefficient b of the value or the coefficient c of the time-consuming score. Based on this principle, the values ​​of a, b, and c can be preliminarily set to ensure that they can better reflect the needs and priorities of specific applications. Since the regression application score is an important indicator for evaluating the performance changes of the model, in different industrial scenarios, in order to further optimize these coefficients, it is necessary to use cross-validation technology to test a variety of different coefficient combinations, and finally select a set of coefficient values ​​that can most sensitively reflect the performance changes of the model. For example, set a to 0.5, b to 0.4, c to 0.1, etc.

[0014] Since the scoring system introduced in the present invention integrates the above three important factors, it effectively overcomes the limitation that traditional evaluation indicators cannot comprehensively and intuitively reflect the performance of the regression model. The calculation formula for regression application scoring is shown in (1):

[0015] RAS=a*(Hit Rate)+b*R 2 +c*(Time Score) (1)

[0016] Among them, a, b, c are three coefficients, Hit Rate is the hit rate within 3% error, R 2 is the determination coefficient, and its calculation formula is shown in formula (2). Time Score is the time-consuming score, and its calculation formula is shown in formula (3).

[0017]

[0018] Among them, yi is the actual value of industrial parameters, is the actual average value of industrial parameters, is the predicted value of industrial parameters, n is the total sample size, x i The time taken to train the model after deleting the i-th feature, x max is the time taken to train the model using all features.

[0019] The process of feature screening in the present invention is as follows Figure 2 As shown in the figure, the feature screening algorithm based on regression application score first uses all features to train the model and records the regression application score, which is temporarily recorded as the baseline score. Then, the algorithm temporarily deletes a feature, uses the remaining features to train the model again and records the new regression application score. If the regression application score of the model is higher than the previous baseline score after temporarily excluding a feature, it means that the feature is not necessary for building the model and should be deleted; conversely, if the regression application score is lower than the previous baseline score, it means that the feature is an important feature required by the model, and deleting it will lead to reduced model performance, so it should be retained. Through this process, the algorithm excludes all features one by one, and finally selects the optimal feature set required for training the model.

[0020] Step 3: Iteratively update the training model to predict the parameters of the next moment

[0021] Finally, the optimal feature set is used to train the random forest model to predict the industrial parameters required at the next time point. At set intervals, the algorithm collects the most recent data set and retrains the model with this new data to optimize the prediction capability of industrial parameters in the future.

[0022] The time interval for model updates and the size of the data set used to train the model need to be set according to the specific industrial field. When the amount of training data is small, the model may be underfitting due to failure to learn sufficiently rich data features, thereby affecting its prediction performance. On the contrary, if the amount of training data is too large, it will not only increase the computational burden, but may also cause overfitting problems, resulting in a decrease in the model's generalization ability for new data. To this end, the present invention first selects 1 day of data as the initial training set to iteratively update and train the model. After each training is completed, the generated model is used to predict the key quality indicators of the following day, and the regression application score, R 2 The hit rate of the value and error is within 3%. Next, the scale of the training set is gradually expanded, and the above process is repeated using 2 days, 3 days...n days (when a data set scale with good results appears, n will not be expanded). That is, after each iteration, the model is updated, the key quality indicators of the next day are predicted and the corresponding evaluation indicators are recorded. This is intended to explore the impact of different training set sizes on model performance and find the most suitable amount of training data. In view of the fact that the inventors have performed multiple rounds of iterations on more than 50,000 data throughout the process, three evaluation indicators are recorded in each round, which makes it difficult to directly compare the performance of training sets of different lengths. Therefore, the present invention calculates the average regression application score, average R 2 The optimal amount of training data is determined by comprehensively evaluating the value and the average hit rate within the error range of 3%. This method helps to overcome the analysis complexity brought by a large amount of data iteration and provides a scientific basis for selecting the optimal training set size.

[0023] By regularly updating the model with the latest data, the effectiveness of the model and the stability of the prediction accuracy are ensured, avoiding the degradation of model performance due to data aging. This continuous learning and self-adjustment mechanism ensures that the model can maintain a high level of prediction accuracy in the long term and adapt to the ever-changing industrial production environment.

[0024] 3. Beneficial effects

[0025] The present invention has the following beneficial effects:

[0026] (1) When the present invention performs data preprocessing, the missing values ​​are filled with the data collected nearby, which can better retain the local characteristics of the data; the isolation forest algorithm is combined to detect and delete outliers, which further improves the purity of the data; in addition, the data that should not be 0 is eliminated through process analysis, which ensures the rationality and accuracy of the data, thereby comprehensively improving the data quality.

[0027] (2) The present invention proposes a feature screening algorithm based on regression application scoring, which integrates three factors for feature screening, effectively making up for the deficiency that traditional evaluation indicators cannot intuitively reflect the performance of regression models; by accurately screening the features required for training the model, the algorithm not only reduces the computational pressure, but also significantly improves the overall performance of the model.

[0028] (3) The present invention proposes a method for iteratively updating the parameters of the training model prediction in the following adjacent period of time, which solves the problem of model timeliness and greatly improves the long-term stability of the model prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of the prediction method of the present invention;

[0030] Figure 2 A flow chart of feature screening in the present invention;

[0031] Figure 3 This is a schematic diagram of the ignition furnace and its component pipelines in the embodiment;

[0032] Figure 4 It is a schematic diagram of the prediction effect in the embodiment. The predicted value and the actual value have a high degree of fit, and a good effect is achieved. DETAILED DESCRIPTION

[0033] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings.

[0034] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0035] The present invention will be further described below in conjunction with the embodiments.

[0036] Example

[0037] Combination Figure 1-Figure 4 As shown, the above-mentioned high-efficiency adaptive complex industrial parameter prediction method based on regression application scoring of the present invention takes the sintering ignition temperature prediction as an example:

[0038] First, the field data is collected through PLC, and the collected data is tested for missing values, and filled by using the values ​​of the nearest neighbor data points to maintain the integrity and consistency of the data set. Then, the isolation forest algorithm is used, the number of decision trees in the forest is set to 120, outlier detection is performed, and the first 1% of the data with the shortest path is deleted. Figure 3 As shown, during sintering ignition, the pressure before the valve, the pressure after the valve, the gas flow, the pressure at the front end of the main pipe and the gas flow of the ignition furnace, these sintering ignition characteristics should not have data with a value of 0. If these characteristics have a value of 0, it means that these data are abnormal and these data samples with a value of 0 need to be deleted.

[0039] Through the above data collection and preprocessing process, a high-quality data set is obtained. Then, the feature screening algorithm based on regression application scoring is used to screen features. The feature screening process is as follows: Figure 2 As shown, the model selected in the screening algorithm is random forest, and the screened features are shown in Table 1 below. Table 2 shows the features screened using the traditional Pearson correlation algorithm. Comparing Table 1 and Table 2, it can be seen that using the feature screening algorithm based on regression application scoring of the present invention, the number of features required to be screened is only 6, which is significantly lower than the number of features screened using the traditional Pearson correlation algorithm, which shows that the algorithm proposed by the present invention greatly reduces the calculation pressure.

[0040] Table 1 Input features selected based on regression application scores

[0041] Feature variable name Feature Abbreviation Blast furnace gas regulating valve opening D02 Gas main rear end pressure D07 Ignition furnace gas flow D09 Ignition furnace gas regulating valve opening D10 Air flow D11 Current temperature of ignition furnace D17

[0042] Table 2 Input features selected by Pearson correlation algorithm

[0043]

[0044]

[0045] After the feature screening algorithm based on regression application scoring was used to screen the sintering ignition data and obtain the best feature combination, the random forest regression prediction model was trained to predict the sintering temperature in the future. The various indicators of the sintering ignition temperature in the next 1 to 5 days predicted by the random forest model are shown in Table 3. It can be seen from Table 3 that the effect of the model becomes worse and worse with the increase of time.

[0046] Table 3. Changes in model prediction effect

[0047] 1 day 2 days 3 days 4 days 5 days Regression App Rating 0.95 0.92 0.87 0.81 0.77 Hit rate within 3% error 98％ 94％ 91％ 84％ 80％ <![CDATA[R 2 ]]> 0.96 0.91 0.85 0.81 0.78 Mean absolute error 6℃ 9℃ 14℃ 17℃ 22℃

[0048] In order to solve the timeliness problem of the model, the present invention proposes a method for regularly iterating and updating the model to predict the sintering ignition temperature in the future. Through multiple experimental verifications, the present invention found that when using the data from the most recent 5 days to train the model to predict the sintering ignition temperature of the next day, not only the model performance is ideal but also the computing power requirements are moderate. The specific operation is: every other day, collect the latest 5 days of data to retrain the model, and then use the newly trained model to predict the sintering ignition temperature of the next day. By continuously repeating this iterative update process, it is ensured that the model can reflect the changes in the latest data in a timely manner, thereby maintaining and improving the accuracy and stability of the prediction. One of the prediction results is as follows: Figure 3 As shown, the predicted value has a high degree of fit with the actual value, and a good effect has been achieved. Table 4 lists the traditional random forest model based on the iterative update training of the Pearson correlation feature screening algorithm, and the random forest model based on the iterative update training of the feature screening algorithm proposed in the present invention to predict the various evaluation indicators of the sintering ignition temperature in the next 30 days. By comparing the various indicators in Table 4, it can be seen that the model trained using the features screened by the algorithm of the present invention performs better in predicting the sintering ignition data in the next 30 days. Specifically, the model has good performance in regression application score, hit rate within 3% error, R 2 And the key indicators such as mean absolute error are better than the model trained only with the Pearson correlation feature screening algorithm. This not only proves that the algorithm of the present invention can effectively improve the accuracy of the model, but also enhances the generalization ability of the model, that is, the adaptability and accuracy of the model on unseen data. Furthermore, when comparing the various indicators of the random forest model iteratively updated and trained based on the regression application scoring feature screening algorithm (see Table 4) with the random forest model based on the regression application scoring feature screening algorithm that is not iteratively updated (see Table 3), it can be clearly observed that the various indicators of the iterative update training model have been significantly improved. This result shows that by iteratively updating the training of the random forest model to predict the ignition temperature at the next time point, the overall performance of the model can be significantly improved, and the stability and reliability of the prediction can be improved.

[0049] Table 4 Comparison of the effects of two feature screening algorithms on training random forest models to predict ignition temperature in the next 30 days

[0050]

[0051] The above is a schematic description of the present invention and its implementation mode, which is not restrictive and is only one of the implementation modes of the present invention, and is not actually limited thereto. Therefore, if a person skilled in the art is inspired by it and, without departing from the purpose of the invention, designs a structure and an implementation mode similar to the technical solution without creativity, they shall all fall within the protection scope of the present invention.

Claims

1. A high-performance adaptive complex industrial parameter prediction method based on regression application scoring, characterized by: The process includes the following: Step 1: Data collection and preprocessing: Install the PLC acquisition equipment at the location to be detected, and preprocess the collected data to fill in missing values ​​and remove outliers; Step 2: Feature screening: Introduce regression application scoring method, combined with key indicators: hit rate within 3% error, determination coefficient R 2 The optimal feature set required for training the model is screened out by averaging the value and time consumption. Step 3. Iteratively update the training model to predict the parameters at the next moment: Use the selected optimal feature set to train the random forest model to predict the industrial parameters required at the next time point; regularly update the model with the latest data to optimize the prediction ability of industrial parameters in the future.

2. According to claim 1, a high-performance adaptive complex industrial parameter prediction method based on regression application scoring is characterized by: The data preprocessing process in step 1 is: fill the missing values ​​with the values ​​of the nearest neighbor data points, and use the isolation forest technique to perform outlier detection on the data. After completing the anomaly scoring of all data points, set the threshold according to the scoring results to delete the outliers.

3. The high-performance adaptive complex industrial parameter prediction method based on regression application scoring according to claim 2 is characterized by: The data preprocessing process in step 1 also includes: after using the isolation forest technique to delete outliers, the data is deeply cleaned to further delete data samples that should not have values ​​of 0 but have values ​​of 0.

4. The high-performance adaptive complex industrial parameter prediction method based on regression application scoring according to claim 1 is characterized by: The calculation process of regression application score in step 2 is as follows: Introduce three key indicators: hit rate within 3% error range, determination coefficient R 2 The hit rate is multiplied by a, R 2 The value is multiplied by b, and the time score is multiplied by c. The sum of these weights is 1. The calculation formula is shown in (1): RAS=a*(Hit Rate)+b*R 2 +c*(Time Score) (1) Among them, a, b, c are three coefficients, Hit Rate is the hit rate within 3% error, R 2 is the determination coefficient, and its calculation formula is shown in formula (2). Time Score is the time-consuming score, and its calculation formula is shown in formula (3): Among them, y i is the actual value of industrial parameters, is the actual average value of industrial parameters, is the predicted value of industrial parameters, n is the total sample size, x i The time taken to train the model after deleting the i-th feature, x max is the time taken to train the model using all features.

5. The high-performance adaptive complex industrial parameter prediction method based on regression application scoring according to claim 4 is characterized by: The process of feature screening in step 2 is as follows: first, use all features to train the model and record the regression application score, which is temporarily recorded as the benchmark score; then, temporarily delete a feature, use the remaining features to train the model again and record the new regression application score; if the regression application score of the model is higher than the previous benchmark score after temporarily excluding a feature, it means that the feature is not necessary for building the model and should be deleted; conversely, if the regression application score is lower than the previous benchmark score, it means that the feature is an important feature required by the model, and deleting it will lead to reduced model performance, so it should be retained; through this process, all features are excluded one by one, and finally the optimal feature set required for training the model is screened out.

Images