Regression model selection method and system for process industry production data flow

Abstract

The application relates to a regression model selection method and system for a process type industrial production data flow, wherein the method comprises the following steps: A1, acquiring each to-be-evaluated regression model and evaluation data sets corresponding to the to-be-evaluated regression model; A2, dividing each evaluation data set into k sub data sets according to a pre-set rule; A3, acquiring model scores of the to-be-evaluated regression model for the k sub data sets; the model scores of the to-be-evaluated regression model are calculated according to evaluation index values of each sub data set; wherein the evaluation index values of each sub data set comprise a corrected error percentage index value, a fitting degree similarity index value, a vector similarity index value and a structure similarity index value; and A4, determining an optimal to-be-evaluated regression model based on the model scores of each to-be-evaluated regression model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of process industry, and in particular to a regression model selection method and system for process industry production data flow. BACKGROUND

[0002] With the wide application of machine learning technology in process industry production process, the performance of mathematical model directly affects the control effect on the production process. In the prior art, the establishment of regression mathematical model mainly relies on human experience to set hyperparameters, and the hyperparameters are continuously adjusted through model evaluation indexes such as MAE, MSE and RMSE as evaluation criteria, and finally the best model is selected as the output result.

[0003] In the process of industrial production, due to the instability of production conditions, the mathematical model is required to have efficient automatic updating capability to adapt to the frequent fluctuations of the working conditions, and the manpower cannot perform a large number of iterative attempts like machines and lacks detection of the continuous effectiveness of the model. SUMMARY

[0004] (I) Technical problems to be solved

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present application provides a regression model selection method and system for process industry production data flow, which solves the technical problems that the regression model is determined by human experience in the prior art, resulting in low efficiency, and that an effective regression model cannot be determined in the prior art.

[0006] (II) Technical solutions

[0007] In order to achieve the above-mentioned purposes, the main technical solutions adopted by the present application include:

[0008] In a first aspect, the present application provides a regression model selection method for process industry production data flow, which comprises:

[0009] A1, obtaining each to-be-evaluated regression model and the evaluation data set corresponding to the to-be-evaluated regression model;

[0010] A2, dividing each evaluation data set into k sub-data sets according to a pre-set rule;

[0011] A3, obtaining the model score of the to-be-evaluated regression model for the k sub-data sets;

[0012] The model score of the to-be-evaluated regression model is calculated from the evaluation index value of each sub-data set; wherein the evaluation index value of each sub-data set includes: a corrected error percentage index value, a fitting degree similarity index value, a vector similarity index value and a structure similarity index value.

[0013] A4, determining an optimal regression model to be evaluated based on a model score of each regression model to be evaluated.

[0014] Preferably,

[0015] Wherein, the evaluation data set corresponding to the regression model to be evaluated comprises N pieces of evaluation data.

[0016] Each piece of evaluation data comprises an independent variable used by the regression model to be evaluated for training, an actual value of a dependent variable corresponding to the independent variable, a predicted value of the dependent variable obtained by the regression model to be evaluated corresponding to the independent variable, and a timestamp representing the time of the piece of evaluation data.

[0017] Preferably, the A2 specifically comprises:

[0018] The evaluation data set is randomly divided into k sub-data sets in chronological order;

[0019] Wherein, k is a pre-set value, and k≥5;

[0020] Wherein, the k sub-data sets corresponding to the evaluation data set satisfy:

[0021] Any sub-data set contains more than 10 pieces of evaluation data;

[0022] Any two sub-data sets are disjoint;

[0023] The number of pieces of evaluation data in the largest sub-data set is less than or equal to twice the number of pieces of evaluation data in the smallest sub-data set.

[0024] Preferably, the A3 specifically comprises:

[0025] A31, for the k sub-data sets, respectively obtaining an evaluation index value of each sub-data set;

[0026] A32, based on the evaluation index value of each sub-data set, obtaining a first evaluation matrix of the regression model to be evaluated corresponding to the k sub-data sets;

[0027] A33, based on the first evaluation matrix of the regression model to be evaluated and a pre-set threshold value, obtaining a second evaluation matrix of the regression model to be evaluated;

[0028] A34, based on the second evaluation matrix of the regression model to be evaluated, obtaining a model score of the regression model to be evaluated.

[0029] Preferably,

[0030] The correction error percentage index value is pre-calculated by formula (1);

[0031] The formula (1) is:

[0032]

[0033] wherein S1(y_true, y_pred) is the modified error percentage index value;

[0034] y_true i is the actual value of the dependent variable in the i-th evaluation data in the sub-data set;

[0035] y_pred i is the predicted value of the dependent variable in the i-th evaluation data in the sub-data set;

[0036] n is the number of evaluation data in the sub-data set;

[0037] The fitting degree similarity index value is calculated in advance by using formula (2);

[0038] The formula (2) is:

[0039] S2(y_true, y_pred) = 1 - |1-w|;

[0040] wherein w is the parameter w corresponding to the sub-data set when the loss value of the loss function is minimized by using the least square method in advance;

[0041] The loss function is:

[0042] wherein E is the loss value of the loss function;

[0043] ε i is the random term in the i-th evaluation data in the sub-data set, which is pre-set to follow the standard normal distribution;

[0044] S2(y_true, y_pred) is the fitting degree similarity index value;

[0045] The vector similarity index value is calculated in advance by using formula (3);

[0046] The formula (3) is:

[0047]

[0048] wherein,

[0049] wherein S3(y_true, y_pred) is the vector similarity index value;

[0050] The structure similarity index value is calculated in advance by using formula (4);

[0051] The formula (4) is:

[0052]

[0053] wherein μ1 is the mean of the actual value y_true of the dependent variable in the sub-data set;

[0054] μ2 is the mean of the predicted value y_pred of the dependent variable in the sub-data set;

[0055] σ1 is the variance of the actual value y_true of the dependent variable in the sub-data set;

[0056] σ2 is the variance of the predicted value y_pred of the dependent variable in the sub-data set;

[0057] σ 12 is the covariance of the actual value y_true of the dependent variable in the sub-data set and the predicted value y_pred of the dependent variable in the sub-data set;

[0058] C1 and C2 are respectively preset constants.

[0059] Preferably,

[0060] The first evaluation matrix is:

[0061]

[0062] wherein the matrix element S 1j represents the modified error percentage index value of the jth sub-data set in the k sub-data sets;

[0063] The matrix element S 2j represents the fitting degree similarity index value of the jth sub-data set in the k sub-data sets;

[0064] The matrix element S 3j represents the vector similarity index value of the jth sub-data set in the k sub-data sets;

[0065] The matrix element S 4j represents the structure similarity index value of the jth sub-data set in the k sub-data sets.

[0066] Preferably, the A33 specifically comprises:

[0067] For any matrix element of the first evaluation matrix, a new value is determined according to the formula (5), and a second evaluation matrix corresponding to the first evaluation matrix is obtained;

[0068] The formula (5) is:

[0069]

[0070] wherein S ij is the element value of the i-th row of the j-th column of the first evaluation matrix, wherein i = 1, 2, 3, 4;

[0071] δ is a pre-set threshold value, and 0.5≥δ≥0.1;

[0072] is a new value of s ij

[0073] The second evaluation matrix is:

[0074]

[0075] Preferably, A34 specifically comprises:

[0076] A341, based on the second evaluation matrix of the regression model to be evaluated, obtaining the number of elements with a value of 0 in the second evaluation matrix of the regression model to be evaluated and the number of elements with a value not equal to 0 in the second evaluation matrix of the regression model to be evaluated;

[0077] A342, based on the number of elements with a value of 0 in the second evaluation matrix of the regression model to be evaluated and the number of elements with a value not equal to 0 in the second evaluation matrix of the regression model to be evaluated, using formula (6) to obtain the model score of the regression model to be evaluated;

[0078] wherein the formula (6) is:

[0079]

[0080] wherein N0 is the number of elements with a value of 0 in the second evaluation matrix of the regression model to be evaluated;

[0081] N1 is the number of elements with a value not equal to 0 in the second evaluation matrix of the regression model to be evaluated;

[0082] score is the model score of the regression model to be evaluated.

[0083] Preferably, the method further comprises:

[0084] A5, obtaining the independent variable corresponding to the optimal regression model to be evaluated and inputting it into the optimal regression model to be evaluated to obtain the corresponding dependent variable.

[0085] On the other hand, the embodiment provides a system for executing the regression model selection method for process-oriented industrial production data flow as described above, and the system comprises:

[0086] a receiving module for obtaining each regression model to be evaluated and the evaluation data set corresponding to the regression model to be evaluated;​

[0087] a dividing module, configured to divide the evaluation data set and divide each of the evaluation data sets into k sub-data sets according to a preset rule;

[0088] a scoring module, configured to obtain a model score of the regression model to be evaluated for the k sub-data sets;

[0089] The model score of the regression model to be evaluated is calculated from an evaluation index value of each sub-data set, wherein the evaluation index value of each sub-data set includes a corrected error percentage index value, a fitting degree similarity index value, a vector similarity index value and a structure similarity index value.

[0090] a determining module, configured to determine an optimal regression model to be evaluated based on the model score of each regression model to be evaluated.

[0091] (Three) beneficial effects

[0092] The regression model selection method and system for process-type industrial production data flow have the following beneficial effects: the regression model to be evaluated and the evaluation data set corresponding to the regression model to be evaluated are obtained; each of the evaluation data sets is divided into k sub-data sets according to a preset rule; the model score of the regression model to be evaluated is obtained for the k sub-data sets; and the optimal regression model to be evaluated is determined based on the model score of each regression model to be evaluated. The model score of the regression model to be evaluated obtained in the present application reflects the evaluation index of the comprehensive performance of the model from different angles, realizes the quantitative scoring of the regression model, saves the manpower and time cost of modeling, and solves the problem that a single evaluation index cannot truly reflect the effectiveness of the regression model. BRIEF DESCRIPTION OF DRAWINGS

[0093] Figure 1 FIG. 1 is a flowchart of a regression model selection method for process-type industrial production data flow according to the present application;

[0094] Figure 2 FIG. 2 is a flowchart of a regression model selection method for process-type industrial production data flow according to the second embodiment of the present application;

[0095] Figure 3 FIG. 3 is a schematic diagram of a regression model determination system for process-type industrial production data flow according to the present application;

[0096] Figure 4 FIG. 4 is a regression model prediction error distribution diagram with a model score of 0.85 according to the present application;

[0097] Figure 5 FIG. 5 is a regression model prediction error distribution diagram with a model score of 0.24 according to the present application. DETAILED DESCRIPTION

[0098] In order to better explain the present application, so as to be understood, the following will be described in detail by specific embodiments, combined with the drawings.

[0099] In order to better understand the above technical solutions, the exemplary embodiments of the present application will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided so that the present application can be more clearly, thoroughly understood, and the scope of the present application can be completely conveyed to those skilled in the art.

[0100] Embodiment one

[0101] Referring to Figure 1 The embodiment provides a regression model selection method for process industry production data flow, and the method comprises the following steps:

[0102] A1, obtaining each to-be-evaluated regression model and an evaluation data set corresponding to the to-be-evaluated regression model.

[0103] The evaluation data set corresponding to the to-be-evaluated regression model comprises N evaluation data. Each evaluation data comprises an independent variable used for training the to-be-evaluated regression model, an actual value of a dependent variable corresponding to the independent variable, a predicted value of the dependent variable obtained by the to-be-evaluated regression model corresponding to the independent variable, and a time stamp representing a time of the evaluation data.

[0104] A2, dividing each evaluation data set into k sub-data sets according to a pre-set rule.

[0105] The A2 specifically comprises:

[0106] The evaluation data set is randomly divided into k sub-data sets in chronological order.

[0107] The k is a pre-set value, and k≥5.

[0108] The k sub-data sets corresponding to the evaluation data set satisfy:

[0109] Any sub-data set contains more than 10 evaluation data, any two sub-data sets are disjoint, and the number of evaluation data in the largest sub-data set is less than or equal to twice the number of evaluation data in the smallest sub-data set.

[0110] A3, obtaining a model score of the to-be-evaluated regression model for the k sub-data sets.

[0111] The model score of the regression model to be evaluated is calculated from the evaluation index value of each sub-data set; wherein the evaluation index value of each sub-data set includes: a corrected error percentage index value, a fitting degree similarity index value, a vector similarity index value, and a structural similarity index value.

[0112] The A3 specifically includes:

[0113] A31, for k sub-data sets, respectively acquiring the evaluation index value of each sub-data set.

[0114] A32, based on the evaluation index value of each sub-data set, acquiring a first evaluation matrix of the regression model to be evaluated corresponding to the k sub-data sets.

[0115] A33, based on the first evaluation matrix of the regression model to be evaluated and a pre-set threshold value, acquiring a second evaluation matrix of the regression model to be evaluated.

[0116] A34, based on the second evaluation matrix of the regression model to be evaluated, acquiring the model score of the regression model to be evaluated.

[0117] A4, based on the model score of each regression model to be evaluated, determining an optimal regression model to be evaluated.

[0118] The regression model selection method and system for process type industrial production data flow in the embodiment, since each regression model to be evaluated and the evaluation data set corresponding to the regression model to be evaluated are acquired; then each evaluation data set is divided into k sub-data sets according to a pre-set rule; and for k sub-data sets, the model score of the regression model to be evaluated is acquired; finally, based on the model score of each regression model to be evaluated, an optimal regression model to be evaluated is determined. The model score of the regression model to be evaluated acquired in the application can reflect the evaluation index of the comprehensive performance of the model from different angles, realize the quantitative score of the regression model, save the manpower and time cost of modeling, and solve the problem that a single evaluation index cannot truly reflect the effectiveness of the regression model.

[0119] Embodiment two

[0120] Referring to Figure 1 , the embodiment provides a regression model selection method for process type industrial production data flow, which comprises:

[0121] A1, acquiring each regression model to be evaluated and the evaluation data set corresponding to the regression model to be evaluated;

[0122] Wherein, the evaluation data set corresponding to the regression model to be evaluated includes: N evaluation data.

[0123] Each piece of evaluation data includes: independent variables used by the regression model to be evaluated, actual values of dependent variables corresponding to the independent variables, predicted values of dependent variables obtained by the regression model to be evaluated corresponding to the independent variables, and a timestamp representing the time of the piece of evaluation data.

[0124] In the actual application of the embodiment, the evaluation data set corresponding to the regression model to be evaluated is shown in Table 1:

[0125] Table 1

[0126]

[0127] Specifically, in the embodiment, the regression model for process-type industrial production data flow can be a boiler output prediction regression model based on boiler operating parameters, and the features used by the regression model are as follows:

[0128] Independent variables: coal amount, air amount, drum pressure, and oxygen content of flue gas.

[0129] Dependent variable: boiler output (main steam flow).

[0130] In the embodiment, the above independent variables (coal amount, air amount, drum pressure, and oxygen content of flue gas) are only one of a plurality of combinations of independent variables. Different combinations of features (independent variables) or regression model fitting methods are used to train a plurality of regression models, and the optimal boiler output prediction regression model is determined by using the regression model selection method for process-type industrial production data flow in the embodiment. The optimal boiler output prediction regression model is used to predict the boiler output and adjust the subsequent operating parameters, so as to achieve the purpose of precise control and safe operation of the boiler.

[0131] A2, dividing each of the evaluation data sets into k sub-data sets according to a pre-set rule; specifically including:

[0132] Randomly dividing the evaluation data set into k sub-data sets in chronological order.

[0133] Wherein, k is a pre-set value, and k≥5.

[0134] Wherein, the k sub-data sets corresponding to the evaluation data set satisfy:

[0135] Any sub-data set contains more than 10 pieces of evaluation data.

[0136] Any two sub-data sets are disjoint.

[0137] The number of pieces of evaluation data in the largest sub-data set is less than or equal to twice the number of pieces of evaluation data in the smallest sub-data set.

[0138] A3, for the k sub-datasets, obtaining the model score of the regression model to be evaluated.

[0139] The model score of the regression model to be evaluated is calculated from the evaluation index value of each sub-dataset; wherein the evaluation index value of each sub-dataset includes: a corrected error percentage index value, a fitting degree similarity index value, a vector similarity index value, and a structure similarity index value.

[0140] In the practical application of the present embodiment, the A3 specifically includes:

[0141] A31, for the k sub-datasets, respectively obtaining the evaluation index value of each sub-dataset.

[0142] The corrected error percentage index value is calculated in advance using formula (1).

[0143] The formula (1) is:

[0144]

[0145] The S1(y_true, y_pred) is the corrected error percentage index value.

[0146] y_true i is the actual value of the dependent variable in the i-th evaluation data in the sub-dataset.

[0147] y_pred i is the predicted value of the dependent variable in the i-th evaluation data in the sub-dataset.

[0148] n is the number of evaluation data in the sub-dataset.

[0149] In the present embodiment, the average value of all actual values y_true in the evaluation data set is used as the denominator in the formula, which can avoid the problems such as singular points with a denominator of 0, very large changes in results caused by small errors, etc. when a series of particularly small denominators y_true appear in the absolute percentage error of the prior art. The corrected error percentage index value in the present embodiment is between [0, 1], and the higher the score, the better the model.

[0150] The fitting degree similarity index value is calculated in advance using formula (2);

[0151] The formula (2) is:

[0152] S2(y_true, y_pred) = 1 - |1 - w|;

[0153] Wherein, w is the parameter w corresponding to the minimum loss value of the loss function when the least squares method is used to solve the minimum loss value of the loss function for the sub-dataset.

[0154] The loss function is:

[0155] Wherein, E is the loss value of the loss function.

[0156] ε i is a random term corresponding to the ith evaluation data in the sub-data set, which is subject to a standard normal distribution.

[0157] S2(y_true, y_pred) is the value of the fitting degree similarity index.

[0158] In this embodiment, it is assumed that there is a linear relationship between the actual value y_true and the predicted value y_pred of the target variable y_true = w*y_pred + ε; wherein ε is a random term subject to a standard normal distribution.

[0159] The least squares method is used to solve the parameter w that minimizes the loss value E of the loss function, and the fitting degree similarity score index S2(y_true, y_pred) = 1 - |w| is constructed. The fitting degree similarity index value is between [0, 1], and the higher the score, the better the model.

[0160] The vector similarity index value is calculated in advance using formula (3).

[0161] The formula (3) is:

[0162]

[0163] Wherein,

[0164] Wherein, S3(y_true, y_pred) is the value of the vector similarity index.

[0165] In this embodiment, the trend of the target variable along the time axis is actually regarded as an n-dimensional vector, and the cosine similarity of the actual value and the predicted value can represent the trend closeness of the two groups of data, and the calculation formula is:

[0166]

[0167] In this embodiment, the cosine similarity is mapped between the interval [0, 1] as the vector similarity score of the actual value and the predicted value:

[0168]

[0169] The structure similarity index value is calculated in advance using formula (4).

[0170] The formula (4) is:

[0171]

[0172] wherein μ1 is the mean of the actual values y_true of the dependent variable in the sub-data set.

[0173] μ2 is the mean of the predicted values y_pred of the dependent variable in the sub-data set.

[0174] σ1 is the variance of the actual values y_true of the dependent variable in the sub-data set.

[0175] σ2 is the variance of the predicted values y_pred of the dependent variable in the sub-data set.

[0176] σ 12 is the covariance of the actual values y_true of the dependent variable in the sub-data set and the predicted values y_pred of the dependent variable in the sub-data set.

[0177] c1 and c2 are constants preset to be close to 0.

[0178] In the embodiment, the structural similarity index value is between [0, 1], and the higher the score is, the better the model is.

[0179] A32, based on the evaluation index value of each sub-data set, obtains a first evaluation matrix of the regression model to be evaluated corresponding to the k sub-data sets.

[0180] The first evaluation matrix is:

[0181]

[0182] wherein the matrix element S 1j represents the modified error percentage index value of the jth sub-data set in the k sub-data sets.

[0183] The matrix element S 2j represents the fitting degree similarity index value of the jth sub-data set in the k sub-data sets.

[0184] The matrix element S 3j represents the vector similarity index value of the jth sub-data set in the k sub-data sets.

[0185] The matrix element S 4j represents the structural similarity index value of the jth sub-data set in the k sub-data sets.

[0186] The regression model selection method for process-oriented industrial production data flow in the embodiment forms a first evaluation matrix capable of reflecting the comprehensive performance of the regression model for process-oriented industrial production data flow from different angles such as the error percentage between the true value and the predicted value of the dependent variable, the fitting degree, and the similarity, realizes quantitative scoring of the regression model for process-oriented industrial production data flow, saves the manpower and time cost of modeling, and solves the problem that a single evaluation index cannot truly reflect the effectiveness of the regression model.

[0187] A33, based on the first evaluation matrix of the regression model to be evaluated and the pre-set threshold value, obtaining the second evaluation matrix of the regression model to be evaluated.

[0188] In the practical application of the embodiment, the A33 specifically comprises:

[0189] For any matrix element of the first evaluation matrix, a new value is determined according to formula (5), and a second evaluation matrix corresponding to the first evaluation matrix is obtained.

[0190] The formula (5) is:

[0191]

[0192] Wherein, s ij is the element value of the i-th row of the j-th column of the first evaluation matrix, and i=1, 2, 3, 4.

[0193] δ is a pre-set threshold value, and 0.5≥δ≥0.1.

[0194] is the new value of s ij .

[0195] The second evaluation matrix is:

[0196]

[0197] A34, based on the second evaluation matrix of the regression model to be evaluated, obtaining the model score of the regression model to be evaluated.

[0198] Specifically, the embodiment A34 specifically comprises:

[0199] A341, based on the second evaluation matrix of the regression model to be evaluated, obtaining the number of elements with a value of 0 in the second evaluation matrix of the regression model to be evaluated and the number of elements with a value not equal to 0 in the second evaluation matrix of the regression model to be evaluated.

[0200] A342、based on the number of elements with a value of 0 in the second evaluation matrix of the regression model to be evaluated and the number of elements with a value other than 0 in the second evaluation matrix of the regression model to be evaluated, the model score of the regression model to be evaluated is obtained by using formula (6).

[0201] The formula (6) is:

[0202]

[0203] N0 is the number of elements with a value of 0 in the second evaluation matrix of the regression model to be evaluated.

[0204] N1 is the number of elements with a value other than 0 in the second evaluation matrix of the regression model to be evaluated.

[0205] score is the model score of the regression model to be evaluated.

[0206] A4, based on the model score of each regression model to be evaluated, the optimal regression model to be evaluated is determined.

[0207] Referring to Figure 2 , the embodiment method further comprises: A5, obtaining the independent variable corresponding to the optimal regression model to be evaluated and inputting it into the optimal regression model to be evaluated to obtain the corresponding dependent variable.

[0208] Referring to Figure 3 , the embodiment further provides a system for executing the regression model selection method for process-oriented industrial production data flow as described above, and the system comprises:

[0209] The receiving module is configured to obtain each regression model to be evaluated and the evaluation data set corresponding to the regression model to be evaluated.

[0210] The division module is configured to divide the evaluation data set into k sub-data sets according to a pre-set rule.

[0211] The scoring module is configured to obtain the model score of the regression model to be evaluated for the k sub-data sets.

[0212] The model score of the regression model to be evaluated is calculated from the evaluation index value of each sub-data set; wherein the evaluation index value of each sub-data set comprises a corrected error percentage index value, a fitting degree similarity index value, a vector similarity index value, and a structure similarity index value.

[0213] The determination module is configured to determine the optimal regression model to be evaluated based on the model score of each regression model to be evaluated.

[0214] As Figure 4 , Figure 5respectively are error distribution diagrams predicted by regression models with scores of 0.85 and 0.24 respectively, the horizontal axis is prediction error, and the vertical axis is error probability density (i.e. error probability), it can be found by comparison that the error distribution interval of the regression model with a higher score is small and the distribution is more concentrated, which shows that the regression model selection method for process industrial production data flow in the embodiment can determine a regression model that meets the real effect of the regression model. Figure 4

[0215] The regression model selection method and system for process industrial production data flow in the embodiment, since each regression model to be evaluated and the evaluation data set corresponding to the regression model to be evaluated are obtained, then each evaluation data set is divided into k sub-data sets according to a pre-set rule, and the model score of the regression model to be evaluated is obtained for the k sub-data sets, and finally the optimal regression model to be evaluated is determined based on the model score of each regression model to be evaluated. The model score of the regression model to be evaluated obtained in the embodiment can reflect the evaluation index of the comprehensive performance of the model from different angles, realize quantitative scoring of the advantages and disadvantages of the regression model, save the manpower and time cost of modeling, and solve the problem that a single evaluation index cannot truly reflect the effectiveness of the regression model.

[0216] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can adopt a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0217] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system) and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be realized by computer program instructions.

[0218] ​It should be noted that the description using "comprises" or "comprising" does not exclude the presence of elements or steps other than those listed in a claim. The words "a" or "an" preceding the disclosure of a plurality of elements or steps do not exclude the presence of a plurality of such elements or steps. The application can be implemented by means of both hardware and software, and any combination thereof. In a claim reciting a means, the term "means" is intended to refer to a combination of the elements in the claim, and is not intended to refer to a specific structure or composition of the combination. The mere fact that certain measures are recited in mutually different claims does not indicate that a combination of these measures cannot be used to advantage. The word "comprising" does not exclude not excluding other elements or steps. The words "first", "second", "third", and the like in the description do not necessarily have an ordinal meaning. These words are used to distinguish between similar elements or steps. The implementation of any of the measuring methods described can be performed by means of any appropriate technique known to the person skilled in the art.

[0219] Furthermore, it is noted that the description of the present application is made in relation to specific embodiments thereof, and that the terms used are to be construed in the context of the specification as a whole and claims appended hereto, and not in an isolated manner. In addition, the terms "a", "an", "the" and the like do not preclude the presence of more than one of the referred items or steps. The terms "first", "second", "third" and the like, merely identify content of the features, states, constituents, materials, components or steps, and do not require a sequential or chronological order. The features, constituents, materials, components or steps can be combined in a plurality of ways and can be used independently, in combination, or in any suitable sub-combination. Furthermore, embodiments described herein can be combined with other embodiments in a suitable manner without departing from the scope of the present application.

[0220] Although preferred embodiments of the application have been described, those skilled in the art will be able to make additional modifications and variations without departing from the scope of the application. Accordingly, the claims should be interpreted to include all such modifications and variations as falling within the scope of the present application.

[0221] It is apparent that a person skilled in the art can make various modifications and variations to the application without departing from the spirit and scope of the application. Accordingly, the application should be construed to include all such modifications and variations as falling within the scope of the claims and equivalents thereof.

Claims

1. A regression model selection method for process-oriented industrial production data streams, characterized by, The method comprises: A1, obtaining each to-be-evaluated regression model and the evaluation data set corresponding to the to-be-evaluated regression model; A2, dividing each evaluation data set into k sub-data sets according to a pre-set rule; A3, obtaining the model score of the to-be-evaluated regression model for the k sub-data sets; The model score of the to-be-evaluated regression model is calculated from the evaluation index value of each sub-data set; wherein the evaluation index value of each sub-data set comprises: a corrected error percentage index value, a fitting degree similarity index value, a vector similarity index value, and a structural similarity index value; The independent variables of the to-be-evaluated regression model include the amount of coal entering the furnace, the amount of air entering the furnace, the pressure of the steam drum, and the oxygen content of the exhaust smoke; The corrected error percentage index value is calculated in advance using formula (1); the formula (1) is: wherein S1(y_true, y_pred) is the modified percentage error index value; y_true i is the actual value of the dependent variable in the i-th evaluation data in the sub-data set; y_pred i is the predicted value of the dependent variable in the i-th evaluation data in the sub-data set; and n is the number of evaluation data in the sub-data set. The fitting degree similarity index value is calculated in advance using formula (2); the formula (2) is: S2(y_true, y_pred) = 1 - |1-w|; wherein w is a parameter corresponding to the sub-data set pre-solved by the least square method to minimize the loss value of the loss function; the loss function is: wherein E is the loss value of the loss function; ε i is a random item corresponding to the i-th evaluation data in the sub-data set and subject to the standard normal distribution; S2(y_true, y_pred) is the fitting degree similarity index value; The vector similarity index value is calculated in advance using formula (3); the formula (3) is: wherein, wherein S3(y true, y pred) is a vector similarity index value; The structural similarity index value is calculated in advance using formula (4); the formula (4) is: wherein μ is the mean of the actual values y true of the dependent variable in the sub-data set; μ2 is the mean of the predicted values y pred of the dependent variable in the sub-data set; σ1 is the variance of the actual values y true of the dependent variable in the sub-data set; σ2 is the variance of the predicted values y pred of the dependent variable in the sub-data set; σ 12 is the covariance of the actual values y true of the dependent variable in the sub-data set and the predicted values y pred of the dependent variable in the sub-data set; c1 and c2 are respectively preset constants; and is the covariance of the actual values y true of the dependent variable in the sub-data set and the predicted values y pred of the dependent variable in the sub-data set; c1 and c2 are respectively preset constants; and A4, determining the optimal to-be-evaluated regression model based on the model score of each to-be-evaluated regression model; A5, predicting the boiler output using the optimal to-be-evaluated regression model to obtain a prediction result.

2. The method of claim 1, Its characteristics are that wherein the evaluation data set corresponding to the to-be-evaluated regression model comprises N evaluation data; Each evaluation data comprises: the independent variables used for training the to-be-evaluated regression model, the actual value of the dependent variable corresponding to the independent variable, the predicted value of the dependent variable obtained by the to-be-evaluated regression model corresponding to the independent variable, and a timestamp representing the time of the evaluation data.

3. The method according to claim 2, c h a r a c t e r i z e d by, The A2 specifically comprises: randomly dividing the evaluation data set into k sub-data sets in chronological order; wherein k is a pre-set value, and k≥5; wherein the k sub-data sets corresponding to the evaluation data set satisfy: any sub-data set contains more than 10 evaluation data; any two sub-data sets are disjoint; the number of evaluation data in the largest sub-data set is less than or equal to twice the number of evaluation data in the smallest sub-data set.

4. The method according to claim 3, c h a r a c t e r i z e d by, The A3 specifically comprises: A31, obtaining the evaluation index value of each sub-data set for the k sub-data sets; A32, obtaining the first evaluation matrix of the to-be-evaluated regression model corresponding to the k sub-data sets based on the evaluation index value of each sub-data set; A33, obtaining the second evaluation matrix of the to-be-evaluated regression model based on the first evaluation matrix of the to-be-evaluated regression model and a pre-set threshold value; A34, obtaining the model score of the to-be-evaluated regression model based on the second evaluation matrix of the to-be-evaluated regression model.

5. The regression model selection method for process-oriented industrial production data flow according to claim 4, characterized in that the first evaluation matrix is: where the matrix elements s 1j denote the modified error percentage indicator value for the jthsub dataset in the k sub datasets. Matrix element s 2j represents the fit similarity indicator value for the jthsub dataset out of k sub datasets. Matrix element s 3j denotes the vector similarity indicator value for the j-th sub dataset out of k sub datasets. Matrix element s 4j denotes the structural similarity index value for the jth sub dataset of the k sub datasets.

6. The method of claim 5, wherein, the A33 specifically comprises: for any matrix element of the first evaluation matrix, a new value is determined according to formula (5), and a second evaluation matrix corresponding to the first evaluation matrix is obtained; The formula (5) is: where s ij is the element value of the ith row of the jth column of the first evaluation matrix, where i = 1, 2, 3, 4; δ is a preset threshold, and 0.5≥δ≥0.1; s ij a new value; The second evaluation matrix is:

7. The method according to claim 6, c h a r a c t e r i z e d by, A34 specifically comprises: A341, based on the second evaluation matrix of the regression model to be evaluated, the number of elements with a value of 0 in the second evaluation matrix of the regression model to be evaluated and the number of elements with a value other than 0 in the second evaluation matrix of the regression model to be evaluated are obtained; A342, based on the number of elements with a value of 0 in the second evaluation matrix of the regression model to be evaluated and the number of elements with a value other than 0 in the second evaluation matrix of the regression model to be evaluated, the model score of the regression model to be evaluated is obtained by using formula (6); Wherein, the formula (6) is: Wherein, N0 is the number of elements with a value of 0 in the second evaluation matrix of the regression model to be evaluated; N1 is the number of elements with a value other than 0 in the second evaluation matrix of the regression model to be evaluated; Score is the model score of the regression model to be evaluated.

8. The method according to claim 7, c h a r a c t e r i z e d by, The method further comprises: A5, obtaining the independent variable corresponding to the optimal regression model to be evaluated and inputting it into the optimal regression model to be evaluated to obtain the corresponding dependent variable.

9. A system for performing a method of selecting a regression model for a data stream of a process-oriented industrial production as claimed in any of claims 1 to 8, characterized in that The system comprises: A receiving module for obtaining each regression model to be evaluated and the evaluation data set corresponding to the regression model to be evaluated; A division module for dividing the evaluation data set into k sub-data sets according to a preset rule; A scoring module for obtaining the model score of the regression model to be evaluated for the k sub-data sets; The model score of the regression model to be evaluated is calculated from the evaluation index value of each sub-data set; wherein, the evaluation index value of each sub-data set comprises: a corrected error percentage index value, a fitting degree similarity index value, a vector similarity index value, and a structure similarity index value; A determination module for determining the optimal regression model to be evaluated based on the model score of each regression model to be evaluated; An execution module for predicting the boiler output by using the optimal regression model to be evaluated to obtain a prediction result.

Images