Manufacturing enterprise ranking evaluation method and system based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM

Abstract

The embodiment of the invention provides a manufacturing enterprise ranking evaluation method and system based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM, and belongs to the field of manufacturing enterprise ranking evaluation methods. The method comprises the steps of obtaining to-be-evaluated index data, and establishing an index system; determining the weight of each index in the index system according to an entropy weight method; according to the determined weight, a TOPSIS method is adopted to obtain a first evaluation value; according to the index system, adopting an SAE method to obtain feature extraction data; obtaining a second evaluation value according to the SOFM method based on the feature extraction data; judging the consistency of the first evaluation value and the second evaluation value; when it is judged that the first evaluation value and the second evaluation value are inconsistent, returning to execute the steps of obtaining the index data and establishing the index system; calculating a first comprehensive evaluation value of the current manufacturing enterprise under the condition that the two are consistent; traversing each manufacturing enterprise to obtain a corresponding first comprehensive evaluation value; and determining the ranking of each manufacturing enterprise according to the first comprehensive evaluation value.

Description

Technical Field

[0001] The present invention relates to the field of manufacturing enterprise ranking evaluation methods, and in particular to a manufacturing enterprise ranking evaluation method and system based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM. Background Art

[0002] In the process of evaluating the competitiveness of manufacturing enterprises, the key links include building an evaluation index system, determining weight coefficients and building a comprehensive evaluation model. First of all, the construction of an evaluation index system requires comprehensive consideration of multiple methods, such as literature analysis, expert consultation, questionnaire survey, Delphi method, grounded theory, etc. They each have their own advantages, but are also affected by subjective factors.

[0003] Secondly, it is the determination of indicator weights. The methods for determining indicator weights mainly include subjective and objective methods. Subjective methods mainly include expert scoring method, hierarchical analysis method, etc., and objective methods mainly include entropy weight method, standard deviation method, CRITIC method, etc. When determining weights by subjective methods, it is inevitable that they will be affected by human factors and produce deviations.

[0004] Finally, it is the construction of a comprehensive evaluation model. Traditional competitiveness evaluation methods include: TOPSIS, DEA, fuzzy comprehensive evaluation method, AHP, GRA. It can be seen that the TOPSIS method is currently one of the most popular methods for multi-criteria decision-making. However, the application of the traditional TOPSIS method has certain limitations, such as being greatly affected by the dimension and being unable to consider the correlation between various indicators, resulting in inaccurate decision results and reduced applicability. Summary of the invention

[0005] The purpose of the embodiment of the present invention is to provide an evaluation method and system for manufacturing enterprise ranking based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM, which solves the problem of inaccurate evaluation results caused by subjective factors when evaluating the ranking of manufacturing enterprises in the prior art.

[0006] In order to achieve the above object, an embodiment of the present invention provides an evaluation method for manufacturing enterprise ranking based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM, the evaluation method comprising:

[0007] Obtain the indicator data to be evaluated and establish an indicator system;

[0008] Determine the weight of each indicator under the indicator system according to the entropy weight method;

[0009] The first evaluation value is obtained by using the TOPSIS method according to the determined weight;

[0010] Acquire feature extraction data using the SAE method according to the indicator system;

[0011] Based on the feature extracted data, a second evaluation value is obtained according to the SOFM method;

[0012] determining the consistency between the first evaluation value and the second evaluation value;

[0013] If it is determined that the first evaluation value and the second evaluation value are inconsistent, returning to the step of obtaining indicator data and establishing an indicator system;

[0014] When it is determined that the first evaluation value and the second evaluation value are consistent, calculating a first comprehensive evaluation value of the current manufacturing enterprise;

[0015] Traversing each of the manufacturing enterprises to obtain the corresponding first comprehensive evaluation value;

[0016] The ranking of each manufacturing enterprise is determined according to the first comprehensive evaluation value.

[0017] Optionally, the evaluation method further includes:

[0018] Obtaining indicator data of each manufacturing enterprise in the previous time period;

[0019] Calculate the ranking of each manufacturing enterprise according to the indicator data as a predicted ranking;

[0020] Obtaining the actual market value ranking of each manufacturing enterprise in the next time period;

[0021] Calculating the error between the predicted ranking and the market value ranking;

[0022] The effectiveness of the evaluation method is determined based on the error.

[0023] Optionally, the obtaining of indicator data to be evaluated and establishing an indicator system includes:

[0024] Establish an initial set of indicators;

[0025] Obtaining preliminary indicator data from the initial indicator set according to the UISHSIC method to construct a final indicator system;

[0026] The step of predicting index data from the initial index set according to the UISHSIC method includes:

[0027] Performing data preprocessing operations on the data in the initial indicator set to obtain a data set;

[0028] Use the dataset as pseudo-labels and select indicators;

[0029] Obtaining the correlation between each of the indicators and the label and the redundancy between each of the indicators and other indicators;

[0030] Screening out the pre-indicator data according to the correlation between the correlation and the redundancy;

[0031] Add the pre-indicator data that has been screened to the indicator sequence until all indicators have been screened;

[0032] Obtain the final sorted pre-indicator data.

[0033] Optionally, the correlation between each of the indicators and the label is obtained according to formula (1).

[0034]

[0035] Among them, HSIC (X i , Y) is the correlation between each indicator and the label, S is all indicators in the data set, u is the total number of samples, X i is the i-th indicator, Y is the label, X (s-i) is the set of remaining indicators of all indicators except indicator i, H is the centralization matrix, and i is the integer code;

[0036] According to formula (2), the redundancy between each of the indicators and other indicators is obtained.

[0037]

[0038] Among them, HSIC (X i , X j ) is the redundancy of the i-th index and the j-th index, X i is the i-th index, X j is the jth indicator;

[0039] According to formula (3), the correlation between the correlation and the redundancy is obtained:

[0040]

[0041] Where Q(T) is the correlation between the correlation and redundancy, σ is the kernel function parameter, ∑ i≠j HSIC(X i , X j ) is the redundancy between the ith indicator and all other indicators.

[0042] Optionally, determining the weight of each indicator under the indicator system according to the entropy weight method includes:

[0043] Establishing an original matrix according to the indicator system;

[0044] Performing data preprocessing on the original matrix according to the range standardization method to obtain a standardized matrix;

[0045] Determining indicator weights according to the standardized matrix;

[0046] According to formula (4), the original matrix is ​​established.

[0047] G={x ab} m×n ,(0< <a<<m,0<<b<<n), (4)

[0048] Among them, m is the number of objects to be evaluated, n is the number of evaluation indicators for each object to be evaluated, a is the object to be evaluated, b is the different evaluation indicators, x ab is the original value of the bth evaluation index of the ath evaluation object, and G is the original matrix;

[0049] The original data is preprocessed according to formula (5) and formula (6), and the standardized matrix is ​​obtained according to formula (7).

[0050]

[0051] E={y ab} m×n , (7)

[0052] Among them, y ab is the standardized value of the bth evaluation index of the ath evaluation object, y ab1 is the normalized value of the positive indicator, y ab2 is the normalized value of the negative indicator, x b is the original value of the bth evaluation index, min(x b ) is the minimum value of the original value of the b-th evaluation index, max(x b ) is the maximum value of the original value of the bth evaluation index, and E is the standardized matrix;

[0053] According to formula (8), the indicator weight is determined:

[0054]

[0055] d b =1-e b , (10)

[0056]

[0057] Among them, z ab is the proportion of the ath evaluation object under the bth indicator, k is the number of evaluation individuals, e b is information entropy, d b is the information redundancy, wb is the indicator weight, w b ∈[0,1],

[0058] Optionally, obtaining the first evaluation value according to the TOPSIS method includes:

[0059] Establish a weighted decision matrix;

[0060] Determining a positive ideal solution and a negative ideal solution according to the weighted decision matrix;

[0061] Determining the distance between each evaluation object and the positive ideal solution and the negative ideal solution;

[0062] Determine a ratio value between each evaluation solution and an ideal solution according to the distance;

[0063] The evaluation schemes are ranked according to the ratio values ​​to determine a first evaluation value.

[0064] Optionally, the weighted decision matrix is ​​determined according to formulas (12) and (13):

[0065] R=(r ab ) m×n , (12)

[0066] r ab =w b ×y ab , (13)

[0067] Among them, R is the weighted decision matrix, r ab is the product of the indicator weight and the standardized value of the bth evaluation indicator of the ath evaluation object;

[0068] According to formulas (14) and (15), the positive ideal solution and the negative ideal solution are determined.

[0069] S + =max(r 1b , r 2b ,…,r mb ), (14)

[0070] S - =min(r 1b , r 2b ,…,r mb ), (15)

[0071] Among them, S + is a positive ideal solution, max(r 1b , r 2b ,…,r mb ) is the maximum value of each column in the weighted decision matrix, S - is a negative ideal solution, min(r1b , r 2b ,…,r mb ) is the minimum value of each column in the weighted decision matrix;

[0072] According to formulas (16) and (17), the distance between each evaluation object and the positive ideal solution and the negative ideal solution is determined.

[0073]

[0074] Among them, D + is the evaluation object and the positive ideal solution S + The distance between - Each evaluation object and the negative ideal solution S - The distance between a =(r a1 ,r a2 ,…,r ab ) is the vector of the ath row in the weighted decision matrix, S -1 is the inverse matrix of the covariance matrix;

[0075] The ratio value is determined according to formula (18),

[0076]

[0077] Among them, C a is a proportional value.

[0078] Optionally, cluster analysis is performed on the feature-extracted data according to formulas (19) and (20).

[0079] c = argmin α ‖γ-w α ‖, (19)

[0080] Among them, c is the best matching neuron, γ is the input data, and w α is the weight vector of all neurons, argmin α ‖γ-w α ‖ is the input data γ and the weight vector w of all neurons α The neuron with the closest distance between them, α is the index of the neuron;

[0081] w α (t+1)=w α (t)+θ(p(t)-w α (t-1)), (20)

[0082] in, For all the best matching neurons α * is the center and all hidden neurons with radius d, θ is the learning rate, w α(t+1), w α (t), w α (t-1) represents the weight vector of the αth neuron at t+1, t, and t-1 time steps, p(t) is the output vector at the tth time step, and t is the time step.

[0083] Optionally, the final comprehensive evaluation value is obtained according to formula (21):

[0084] P T =0.5P1+0.5P2, (21)

[0085] Among them, P T is the comprehensive evaluation value, P1 is the first evaluation value, and P2 is the second evaluation value.

[0086] On the other hand, the present invention also includes an evaluation system for ranking manufacturing enterprises based on UISHSIC-Entropy Weight-TOPSIS and SAE-SOFM, and the evaluation system includes any of the evaluation methods described above.

[0087] Through the above technical scheme, the present invention proposes a comprehensive evaluation method that integrates the unsupervised Hilbert-Schmidt independence criterion (UISHSIC), the entropy weight-ideal solution similarity order preference technique (TOPSIS) method, and the sparse autoencoder (SAE) and self-organizing feature map (SOFM) combined model, which can improve the objectivity and comprehensiveness of the evaluation. First, the present invention uses the UISHSIC method to objectively screen the initial large set of indicators to ensure that the selected indicators are both representative and independent of each other, effectively eliminate redundant information, and establish an indicator system; then, the entropy weight method is used to determine the weight of each key indicator to reflect its importance, and combined with the TOPSIS method based on Mahalanobis distance, a multi-indicator comprehensive evaluation model is constructed to obtain the first evaluation value. Further, by using the data dimensionality reduction and feature extraction capabilities of SAE, combined with the clustering and visualization advantages of SOFM, the competitiveness of manufacturing enterprises is deeply analyzed to obtain the second evaluation value. Finally, based on the consistency of the first evaluation value and the second evaluation value, the final comprehensive evaluation value is obtained to determine the ranking of each manufacturing enterprise.

[0088] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0090] Figure 1is a schematic diagram of an evaluation method for ranking manufacturing enterprises based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to one embodiment of the present invention;

[0091] Figure 2 is a schematic diagram of an evaluation method for ranking manufacturing enterprises based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to one embodiment of the present invention;

[0092] Figure 3 It is a schematic diagram of obtaining the index data to be evaluated and establishing the index system from an evaluation method for manufacturing enterprise ranking based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to one embodiment of the present invention;

[0093] Figure 4 It is a schematic diagram of determining a final indicator system in an evaluation method for ranking manufacturing enterprises based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to an embodiment of the present invention;

[0094] Figure 5 It is a schematic diagram of determining the weight of each indicator under the indicator system in an evaluation method for manufacturing enterprise ranking based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to one embodiment of the present invention;

[0095] Figure 6 It is a schematic diagram of determining a first evaluation value in an evaluation method for manufacturing enterprise ranking based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to an embodiment of the present invention;

[0096] Figure 7 It is a SAE structure diagram in an evaluation method for ranking manufacturing enterprises based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to an embodiment of the present invention;

[0097] Figure 8 A schematic diagram of an initial indicator set for establishing an indicator system in a manufacturing enterprise ranking based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to an embodiment of the present invention;

[0098] Fig. 9 A schematic diagram of a final indicator set for establishing an indicator system in a manufacturing enterprise ranking based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to an embodiment of the present invention;

[0099] Fig.10This is a parameter setting of SAE-SOFM in a manufacturing enterprise ranking based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to an embodiment of the present invention. DETAILED DESCRIPTION

[0100] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0101] In the embodiments of the present application, unless otherwise specified, directional words such as "up, down, top, bottom" are usually used with reference to the directions shown in the drawings or are used to describe the relative positional relationships of components in the vertical, perpendicular or gravity direction.

[0102] In addition, if there are descriptions involving "first", "second", etc. in the implementation methods of this application, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various implementation methods can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0103] like Figure 1 As shown, Figure 1 is a schematic diagram of an evaluation method for ranking manufacturing enterprises based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM according to an embodiment of the present invention, comprising Figure 1 It can be seen that the evaluation method includes:

[0104] In step S01, the indicator data to be evaluated is obtained and an indicator system is established.

[0105] In step S02, the weight of each indicator in the indicator system is determined according to the entropy weight method.

[0106] In step S03, a first evaluation value is obtained by using the TOPSIS method according to the determined weight.

[0107] In step S04, the SAE method is used to obtain feature extraction data according to the indicator system.

[0108] In step S05 , based on the feature extracted data, a second evaluation value is acquired according to the SOFM method.

[0109] In step S06 , the consistency between the first evaluation value and the second evaluation value is determined.

[0110] In step S07, when it is determined that the first evaluation value and the second evaluation value are inconsistent, the process returns to the step of obtaining indicator data and establishing an indicator system.

[0111] In step S08, when it is determined that the first evaluation value and the second evaluation value are consistent, a first comprehensive evaluation value of the current manufacturing enterprise is calculated.

[0112] In step S09, each manufacturing enterprise is traversed to obtain a corresponding first comprehensive evaluation value.

[0113] In step S10, the ranking of each manufacturing enterprise is determined according to the first comprehensive evaluation value.

[0114] In the prior art, the methods for determining the weights of indicators mainly include subjective and objective methods. Subjective methods mainly include expert scoring method, hierarchical analysis method, etc., and objective methods mainly include entropy weight method, standard deviation method, CRITIC method, etc. When the subjective method determines the weight, it is inevitable that it will be affected by human factors and produce deviations. Therefore, in one embodiment of the present invention, an objective method is used to determine the weight of the indicator. In step S02 and step S03, the entropy weight method is used to determine the weight of the indicator, and the first evaluation value is obtained by the TOPSIS method, which can ensure the fairness and scientificity of the evaluation result; in step S04 and step S05, the SAE method and SOFM method are used to obtain the second evaluation value. In the process of extracting data features, the sparse autoencoder will naturally assign different weights to different input indicators, avoiding the subjectivity and deviation caused by artificial weighting. In addition, the sparse autoencoder can more effectively extract representative features by adding sparsity constraints, thereby improving the generalization ability and robustness of the model.

[0115] In this implementation, in order to verify the effectiveness of the evaluation method, in one embodiment of the present invention, as Figure 2 As shown, the evaluation method also includes:

[0116] In step S11, the indicator data of each manufacturing enterprise in the previous time period is obtained.

[0117] In step S12, the ranking of each manufacturing enterprise is calculated based on the indicator data as a predicted ranking.

[0118] In step S13, the actual market value ranking of each manufacturing enterprise in the next time period is obtained.

[0119] In step S14, the error between the predicted ranking and the market value ranking is calculated.

[0120] In step S15, the effectiveness of the evaluation method is determined based on the error.

[0121] In this embodiment, there are many ways to obtain the indicator data to be evaluated and establish the indicator system known to those skilled in the art. In one embodiment of the present invention, Figure 3 As shown, the method of obtaining the indicator data to be evaluated and establishing the indicator system includes the following steps:

[0122] In step S16, an initial indicator set is established.

[0123] In step S17, pre-index data is obtained from the initial index set according to the UISHSIC method to construct a final index system.

[0124] There are many ways known to those skilled in the art for obtaining pre-index data from the initial index set and constructing the final index system according to the UISHSIC method. In one embodiment of the present invention, Figure 4 As shown in the figure, how to obtain the preliminary indicator data from the initial indicator set according to the UISHSIC method and construct the final indicator system includes the following steps:

[0125] In step S171, a data preprocessing operation is performed according to the data in the initial indicator set to obtain a data set;

[0126] In step S172, the data set is used as a pseudo label and an indicator is selected;

[0127] In step S173, the correlation between each indicator and the label and the redundancy between each indicator and other indicators are obtained;

[0128] In step S174, the pre-index data is screened out according to the correlation between the correlation and the redundancy;

[0129] In step S175, the pre-index data selected is added to the index sequence until all the indexes are selected;

[0130] In step S176, the final sorted pre-index data is obtained.

[0131] In the prior art, the construction of the evaluation index system requires comprehensive consideration of multiple methods, such as literature analysis, expert consultation, questionnaire survey, Delphi method, grounded theory, etc., each of which has its own advantages, but is also affected by subjective factors. Therefore, in the present invention, on the basis of using literature analysis to construct the initial index set, the unsupervised UISHSIC method is introduced. According to this method, the key features that can reveal important relationships and classification information in the data and screen out non-redundant key features can be selected from the data to form the final index set, such as Figure 8 and Fig. 9 As shown, Figure 8 is the initial indicator set, Fig. 9 is the final indicator set, Figure 8 and Fig. 9 It can be seen that this evaluation method makes the constructed indicator system more objective.

[0132] In the present invention, based on the construction of the initial indicator set by literature analysis, the unsupervised UISHSIC method is introduced, wherein the Hilbert-Schmidtindependence criterion (HSIC) is an independence metric based on a kernel function, which is known for its advantages such as simple calculation, rapid convergence and low deviation. However, in the prior art, HSIC is mainly used in supervised indicator selection tasks, that is, scenarios where the data has been labeled with labels (Y) in advance. HSIC obtains the correlation judgment criterion by calculating the empirical estimate of the Hilbert-Schmid covariance operator norm between variables on the reproducing kernel Hilbert space. In the present invention, the established indicator system is an unsupervised indicator selection problem (Unsupervised Indicator Selection, UIS), that is, the data is not accompanied by a clear label. Therefore, in one embodiment of the present invention, the entire sample set X is regarded as a pseudo-label of the data, that is, it is assumed that Y=X. In this way, the modified HSIC can be used to calculate the correlation between each indicator and this pseudo-label in the current unsupervised situation, as well as the redundancy between them and other indicators. Such processing enables the present invention to evaluate and screen out the most critical and effective indicators for subsequent analysis in an unsupervised environment.

[0133] In one embodiment of the present invention, how to obtain the correlation between each indicator and the label and the redundancy between each indicator and other indicators can be various methods known to those skilled in the art. In one embodiment of the present invention, the correlation is determined according to formula (1):

[0134]

[0135] Among them, HSIC (X i , Y) is the correlation between each indicator and the label, S is all indicators in the data set, u is the total number of samples, X i is the i-th index, Y is the label vector, X (s-i) is the set of remaining indicators after removing indicator i from all indicators, H is the centralization matrix, and i is the integer code.

[0136] In formula (1), Y=X, Y, X∈R u*u is a string containing element Y fe =y(yf ,y e ) and X fe =x(x f ,x e ), the kernel matrix y f For row f, y e is the e-th column, x f For the fth row, x e Column e, R u*u is a subset matrix, where σ is the kernel function parameter. H=I u -u -1 11 T ∈R u*u is a centralization matrix, 1∈R u is a column vector whose u-dimensional elements are all 1, I u is the identity matrix of u*u, u -1 is the reciprocal of u, where u is an integer code, 1 T is the transposed matrix of the u-dimensional column vector whose elements are all 1.

[0137] Determine the redundancy according to formula (2):

[0138]

[0139] Among them, HSIC (X i , X j ) is the redundancy of the i-th index and the j-th index, X i is the i-th index, X j is the jth indicator. If the i-th indicator is highly correlated with the j-th indicator, then HSIC(X i , X j ) is larger.

[0140] According to formula (3), the correlation between relevance and redundancy is determined.

[0141]

[0142] Among them, Q(T) is the correlation between relevance and redundancy, σ is the kernel function parameter, ∑ i≠j HSIC(X i , X j ) is the redundancy between the ith indicator and all other indicators. From formula (3), we can see that HSIC(X i , the larger the value of Y), the higher the correlation between each indicator and the label, and the higher the correlation between the remaining labels and the label; ∑ i≠j HSIC(X i , X j ) is larger, that is, the redundancy between the i-th indicator and all other indicators is higher.

[0143] In this implementation, the method of determining the weight of each indicator under the indicator system according to the entropy weight method can be a variety of methods known to those skilled in the art. In one embodiment of the present invention, Figure 5 As shown, the method of determining the weight of each indicator under the indicator system according to the entropy weight method includes the following steps:

[0144] In step S18, an original matrix is ​​established according to the indicator system;

[0145] In step S19, the original matrix is ​​preprocessed according to the range standardization method to obtain a standardized matrix;

[0146] In step S20, the indicator weights are determined according to the normalization matrix.

[0147] According to formula (4), the original matrix is ​​established.

[0148] G={x ab} m×n ,(0<<a<<m,0<<b<<n), (4)

[0149] Among them, m is the number of objects to be evaluated, n is the number of evaluation indicators for each object to be evaluated, a is the object to be evaluated, b is the different evaluation indicators, x ab is the original value of the bth evaluation index of the ath evaluation object, and G is the original matrix.

[0150] The original data is preprocessed according to formula (5) and formula (6), and the standardized matrix is ​​obtained according to formula (7).

[0151]

[0152] E={y ab} m×n , (7)

[0153] Among them, y ab is the standardized value of the bth evaluation index of the ath evaluation object, y ab1 is the normalized value of the positive indicator, y ab2 is the normalized value of the negative indicator, x b is the original value of the bth evaluation index, min(x b ) is the minimum value of the original value of the b-th evaluation index, max(x b ) is the maximum value of the original value of the bth evaluation index, and E is the standardized matrix;

[0154] According to formula (8), the indicator weight is determined:

[0155]

[0156] d b =1-e b , (10)

[0157]

[0158] Among them, z ab is the proportion of the ath evaluation object under the bth indicator, k is the number of evaluation individuals, e b is information entropy, d b is the information redundancy, w b is the indicator weight, w b ∈[0,1],

[0159] In this implementation, the method for obtaining the first evaluation value according to the TOPSIS method may be a variety of methods known to those skilled in the art. In one embodiment of the present invention, Figure 6 As shown, the first evaluation value is obtained according to the TOPSIS method, including:

[0160] In step S21, a weighted decision matrix is ​​established;

[0161] In step S22, a positive ideal solution and a negative ideal solution are determined according to the weighted decision matrix;

[0162] In step S23, the distance between each evaluation object and the positive ideal solution and the negative ideal solution is determined;

[0163] In step S24, the ratio value of each evaluation scheme to the ideal solution is determined according to the distance;

[0164] In step S25, the evaluation schemes are ranked according to the ratio values ​​to determine a first evaluation value.

[0165] Among them, the working principle of the TOPSIS method is to set the positive ideal solution (PIS) and the negative ideal solution (NIS) of the multi-criteria decision-making problem, and consider the distance between the evaluated alternatives and PIS and NIS as the evaluation criteria, and sort all the alternatives. If a candidate has the shortest distance to the positive ideal solution and the farthest distance to the negative ideal solution, then it is the best solution and should be selected. Therefore, traditional TOPSIS is a technology based on distance measurement. Traditional TOPSIS generally uses Euclidean distance to calculate the distance between the evaluated solution and the positive and negative ideal solutions, but does not consider the correlation between features. The Mahalanobis distance nonlinearly associates the components inside the evaluation matrix through its own covariance matrix, solves the problem of uneven dimensions of each indicator, and eliminates the related interference between each feature. Because this article uses a sparse autoencoder to extract features for the established indicator system, the features finally extracted are not necessarily unrelated to each other, so the present invention uses the TOPSIS method improved based on the Mahalanobis distance to evaluate the ranking of manufacturing enterprises. In one embodiment of the present invention, the Mahalanobis distance is first used to calculate the distance between each enterprise and the positive ideal solution (PIS) and the negative ideal solution (NIS), and then the competitiveness levels of all enterprises are ranked based on these adaptive distances.

[0166] In one embodiment of the present invention, the weighted decision matrix is ​​determined according to formulas (12) and (13):

[0167] R=(r ab ) m×n , (12)

[0168] r ab =w b ×y ab , (13)

[0169] Among them, R is the weighted decision matrix, r ab is the product of the indicator weight and the standardized value of the bth evaluation indicator of the ath evaluation object;

[0170] According to formulas (14) and (15), the positive ideal solution and the negative ideal solution are determined.

[0171] S + =max(r 1b , r 2b ,…,r mb ), (14)

[0172] S - =min(r 1b , r 2b ,…,r mb ), (15)

[0173] Among them, S +is a positive ideal solution, max(r 1b , r 2b ,…,r mb ) is the maximum value of each column in the weighted decision matrix, S - is a negative ideal solution, min(r 1b , r 2b ,…,r mb ) is the minimum value of each column in the weighted decision matrix;

[0174] According to formulas (16) and (17), the distance between each evaluation object and the positive ideal solution and the negative ideal solution is determined.

[0175]

[0176] Among them, D + is the evaluation object and the positive ideal solution S + The distance between - Each evaluation object and the negative ideal solution S - The distance between a =(r a1 ,r a2 ,…,r ab ) is the vector of the ath row in the weighted decision matrix, S -1 is the inverse matrix of the covariance matrix;

[0177] Determine the ratio value according to formula (18):

[0178]

[0179] Among them, C a This evaluation method ranks companies according to their relative proximity.

[0180] Autoencoder (AE) is a nonlinear feature extraction model for unsupervised learning. AE is implemented in the form of a neural network and is used to reconstruct its input signal. Because the training process of AE does not require data labels, it is an unsupervised learning model and method. The multi-hidden layer structure of AE can learn more complex data. The latter layer of the network learns the data features of the previous layer. The more hidden layers there are, the higher-level data features can be learned, so that more complex data can be processed. It consists of two parts: an encoder and a decoder. The main task of the encoder is to compress the input data, extract the main features, and convert high-dimensional data into low-dimensional representations; the main task of the decoder is to reconstruct the original data from the low-dimensional representation to ensure that the features extracted by the encoder can effectively describe the original data.

[0181] A complete AE consists of three different layers, namely the input layer, hidden layer, and output layer. The number of neurons in each layer is n, m, and n, respectively, where n and m are integer codes. The number of neurons in the input layer and the output layer is the same, and the number of neurons in the hidden layer is unlimited. When the number of neurons in the hidden layer is less than the number of neurons in the input layer, it is called a sparse structure, also known as a compressed structure.

[0182] like Figure 7 As shown, sparse autoencoder (SAE) means that the hidden layer features have sparse response characteristics, and its encoding features can better reconstruct the network input. Introducing sparsity in the DAE standard model makes the input data structure of the network clearer and simplifies the calculation. In one embodiment of the present invention, there are two advantages of using the SAE method: first, compared with methods such as principal component analysis (PCA), SAE not only extracts a part of the indicators with the largest contribution rate to analyze the global situation, but also considers other indicators with lower contribution rates, which is comprehensive; second, the sparse autoencoder will naturally assign different weights to different input indicators, avoiding the subjectivity and deviation caused by artificial weighting. Therefore, the present invention adopts the SAE method to obtain feature extraction data according to the indicator system.

[0183] The Self-Organizing Feature Map (SOFM) network is a competitive neural network with an unsupervised learning mechanism. It uses the idea of ​​dimensionality reduction to map high-dimensional input data to a low-dimensional space through nonlinear transformation, while retaining the topological structure and visualization of the data. Through analysis, we can evaluate the competitiveness of manufacturing companies based on the distribution of data points at each neuron position. Multiple data points clustered on the same neuron indicate that companies have similar performance in competitiveness, while a small number of data points may indicate that companies have unique competitive advantages or disadvantages. This approach provides intuitive data insights to help decision makers understand and compare the competitiveness levels between different companies.

[0184] In the network, neurons in the output layer (competition layer) compete for the input data to fight for the opportunity to respond to the input pattern. In the end, only one neuron will become the winner of the competition, that is, the winning neuron. That is, it mainly uses the competition mechanism to self-organize the data so that the spatial distribution density of the connection weights tends to be consistent with the probability distribution of the input data, that is, the spatial distribution of the connection weight vector can reflect the statistical characteristics of the input data.

[0185] In SOFM, each neuron has a weight vector whose dimension is the same as the dimension of the input data. Assume that the input data is k∈R n , R n is the weight vector of all neurons, then the weight vector of each neuron is w α , w α ∈R n, where α is the index of the neuron. And the weight vector w α It can be seen as a "representation" of the position of this neuron in high-dimensional space. This position will be gradually adjusted as training progresses to better match the distribution of input data.

[0186] According to formulas (19) and (20), cluster analysis is performed on the feature extracted data.

[0187] c = argmin α ‖γ-w α ‖, (19)

[0188] Among them, c is the best matching neuron, γ is the input data, and w α is the weight vector of all neurons, argmin α ‖γ-w α ‖ is the input data γ and the weight vector w of all neurons α The neuron with the closest distance between them is α, which is the index of the neuron.

[0189] SOFM selects the best matching neuron (i.e., winning neuron) for each input data through a competitive mechanism. Specifically, SOFM calculates the weight vector w of the input data k and all neurons α The distance between them is the distance between them, and the neuron with the closest distance is selected as the winning neuron. Here, the weight vector w of the winning neuron is c It is the best "representative" of the current input data.

[0190] w α (t+1)=w α (t)+θ(p(t)-w α (t-1)), (20)

[0191] in, For all the best matching neurons α * is the center and all hidden neurons with radius d, θ is the learning rate, w α (t+1), w α (t), w α (t-1) represents the weight vector of the αth neuron at t+1, t, and t-1 time steps, p(t) is the output vector at the tth time step, and t is the time step.

[0192] The Kohonen rule (update formula) in SOFM makes the weight vector w of each input data point k α It will move closer to the input data, especially the weight vectors of the winning neuron and the neurons in its neighborhood will be adjusted.

[0193] Through multiple iterations, training is performed on all input data, gradually adjusting the weight vectors of all neurons so that they can better represent the distribution of the input data.

[0194] After training, SOFM maps high-dimensional data to low-dimensional space, and the weight vector w of each neuron is α It represents a "center point" of the data in low-dimensional space. For each input data point, find the neuron with the nearest weight vector, and use the index of this neuron to determine the cluster label of the data point. All input data points belonging to the same neuron are classified into the same cluster category.

[0195] In this embodiment, if Fig.10 As shown in the figure, the input of the SAE model is 37 indicator data in the manufacturing enterprise ranking evaluation index system. After encoding and decoding, 37-dimensional feature data is finally output. However, this is not a simple copy of the original input data, but an abstraction and reconstruction of these data at a higher level. These feature data not only retain the key information in the original data, but also integrate new information dimensions that are more critical to competitiveness assessment through the learning process of the model, so that the output data can provide more accurate and in-depth insights when used for subsequent analysis or decision support.

[0196] The output of the SAE model is used as the input of the SOFM network, and the output of the SOFM network is 5-dimensional data. This 5-dimensional data corresponds to the clustering of 28 manufacturing enterprises into five categories. By mapping the 37-dimensional data output by the SAE model with the 7 first-level indicators of the evaluation index system one by one, the performance of each enterprise on each first-level indicator can be obtained. After comprehensive weighting, the comprehensive ranking of enterprise competitiveness can be obtained. The 37-dimensional data output by the SAE model is connected to the 7 first-level evaluation indicators covering multiple aspects of enterprise operations through a mapping mechanism to extract the specific performance level of each manufacturing enterprise on each first-level indicator. Subsequently, the performance under these first-level indicators is summed up, and finally the comprehensive score and ranking of enterprise competitiveness are obtained, that is, the second evaluation value is obtained. This method can comprehensively and objectively reflect the competitiveness of each enterprise in the industry.

[0197] In this embodiment, the manner of determining the consistency between the first evaluation value and the second evaluation value may be any of a variety of methods known to those skilled in the art. In one embodiment of the present invention, the Spearman correlation coefficient is used to test the consistency between the entropy weight-TOPSIS comprehensive evaluation result and the SAE-SOFM evaluation result, that is, the consistency between the first evaluation value and the second evaluation value. The Spearman correlation coefficient is a non-parametric indicator for measuring the dependency of two variables, and is particularly useful in evaluating ranking correlation. Its value ranges from -1 to 1, where 1 indicates a perfect positive correlation (i.e., the two rankings are completely consistent); 0 indicates no correlation (i.e., there is no association between the rankings); and -1 indicates a perfect negative correlation (i.e., one ranking is a complete reversal of another ranking).

[0198] If the results of the two methods are highly consistent, it means that the competitiveness evaluation model constructed is stable and reliable, and can fully reflect the real competitiveness level of the enterprise. On this basis, further in-depth analysis of the enterprise's competitiveness can provide strong support for the enterprise's planning and decision-making. If the results of the two methods are inconsistent, it is necessary to re-examine the selection of evaluation indicators, weight distribution, data processing and other links, find possible reasons and make adjustments until a consistent and reasonable evaluation result is obtained.

[0199] In this embodiment, when the results of the first evaluation value and the second evaluation value are consistent, the final comprehensive evaluation value is obtained according to formula (21):

[0200] P T =0.5P1+0.5P2, (21)

[0201] Among them, P T is the comprehensive evaluation value, P1 is the first evaluation value, and P2 is the second evaluation value.

[0202] In addition, the present invention also includes an evaluation system for ranking manufacturing enterprises based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM, and the evaluation system includes any of the above evaluation methods.

[0203] Through the above technical scheme, the present invention proposes a comprehensive evaluation method that integrates the unsupervised Hilbert-Schmidt independence criterion (UISHSIC), the entropy weight-ideal solution similarity order preference technique (TOPSIS) method, and the sparse autoencoder (SAE) and self-organizing feature map (SOFM) combined model, which can improve the objectivity and comprehensiveness of the evaluation. First, the present invention uses the UISHSIC method to objectively screen the initial large set of indicators to ensure that the selected indicators are both representative and independent of each other, effectively eliminate redundant information, and establish an indicator system; then, the entropy weight method is used to determine the weight of each key indicator to reflect its importance, and combined with the TOPSIS method based on Mahalanobis distance, a multi-indicator comprehensive evaluation model is constructed. Furthermore, by utilizing the data dimensionality reduction and feature extraction capabilities of SAE, combined with the clustering and visualization advantages of SOFM, the competitiveness of manufacturing enterprises is deeply analyzed.

[0204] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

[0205] The optional implementation modes of the examples of the present invention are described in detail above in conjunction with the accompanying drawings; however, the implementation modes of the present invention are not limited to the specific details in the above implementation modes; within the technical concept of the implementation modes of the present invention, a variety of simple modifications can be made to the technical solutions of the implementation modes of the present invention, and these simple modifications all belong to the protection scope of the implementation modes of the present invention.

[0206] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0207] In addition, various different embodiments of the embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. An evaluation method for ranking manufacturing enterprises based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM, characterized in that: The evaluation methods include: Obtain the indicator data to be evaluated and establish an indicator system; Determine the weight of each indicator under the indicator system according to the entropy weight method; The first evaluation value is obtained by using the TOPSIS method according to the determined weight; Acquire feature extraction data using the SAE method according to the indicator system; Based on the feature extracted data, a second evaluation value is obtained according to the SOFM method; determining the consistency between the first evaluation value and the second evaluation value; If it is determined that the first evaluation value and the second evaluation value are inconsistent, returning to the step of obtaining indicator data and establishing an indicator system; When it is determined that the first evaluation value and the second evaluation value are consistent, calculating a first comprehensive evaluation value of the current manufacturing enterprise; Traversing each of the manufacturing enterprises to obtain the corresponding first comprehensive evaluation value; The ranking of each manufacturing enterprise is determined according to the first comprehensive evaluation value.

2. The evaluation method according to claim 1, characterized in that: The evaluation method also includes: Obtaining indicator data of each manufacturing enterprise in the previous time period; Calculate the ranking of each manufacturing enterprise according to the indicator data as a predicted ranking; Obtaining the actual market value ranking of each manufacturing enterprise in the next time period; Calculating the error between the predicted ranking and the market value ranking; The effectiveness of the evaluation method is determined based on the error.

3. The evaluation method according to claim 1, characterized in that: The obtaining of the indicator data to be evaluated and the establishment of the indicator system include: Establish an initial set of indicators; Obtaining preliminary indicator data from the initial indicator set according to the UISHSIC method to construct a final indicator system; The step of predicting index data from the initial index set according to the UISHSIC method includes: Performing data preprocessing operations on the data in the initial indicator set to obtain a data set; Use the dataset as pseudo-labels and select indicators; Obtaining the correlation between each of the indicators and the label and the redundancy between each of the indicators and other indicators; Screening out the pre-indicator data according to the correlation between the correlation and the redundancy; Add the pre-indicator data that has been screened to the indicator sequence until all indicators have been screened; Obtain the final sorted pre-indicator data.

4. The evaluation method according to claim 3, characterized in that: Obtaining the correlation between each of the indicators and the label and the redundancy between each of the indicators and other indicators includes: According to formula (1), the correlation between each indicator and the label is obtained. Among them, HSIC (X i , Y) is the correlation between each indicator and the label, S is all indicators in the data set, u is the total number of samples, X i is the i-th indicator, Y is the label, X (s-i) is the set of remaining indicators after removing indicator i from all indicators, H is the centralization matrix, and i is the integer code; According to formula (2), the redundancy between each of the indicators and other indicators is obtained. Among them, HSIC (X i , X j ) is the redundancy of the i-th index and the j-th index, X i is the i-th index, X j is the jth indicator; The pre-indicator data selected according to the correlation between the correlation and the redundancy include: According to formula (3), the correlation between the correlation and the redundancy is obtained. Where Q(T) is the correlation between the correlation and redundancy, σ is the kernel function parameter, ∑ i≠j HSIC(X i , X j ) is the redundancy between the ith indicator and all other indicators.

5. The evaluation method according to claim 1, characterized in that: Determining the weight of each indicator under the indicator system according to the entropy weight method includes: Establishing an original matrix according to the indicator system; Performing data preprocessing on the original matrix according to the range standardization method to obtain a standardized matrix; Determining indicator weights according to the standardized matrix; According to the indicator system, an original matrix is ​​established, including: According to formula (4), the original matrix is ​​established. G={x ab } m×n ,(0<<a<<m,0<<b<<n), (4) Among them, m is the number of objects to be evaluated, n is the number of evaluation indicators for each object to be evaluated, a is the object to be evaluated, b is the different evaluation indicators, x ab is the original value of the bth evaluation index of the ath evaluation object, and G is the original matrix; The original matrix is ​​preprocessed according to the range standardization method to obtain a standardized matrix including: The original data is preprocessed according to formula (5) and formula (6), and the standardized matrix is ​​obtained according to formula (7). E={and ab } m×n , (7) Among them, y ab is the standardized value of the bth evaluation index of the ath evaluation object, y ab1 is the normalized value of the positive indicator, y ab2 is the normalized value of the negative indicator, x b is the original value of the bth evaluation index, min(x b ) is the minimum value of the original value of the b-th evaluation index, max(x b ) is the maximum value of the original value of the bth evaluation index, and E is the standardized matrix; Determining the indicator weights according to the standardized matrix includes: According to formula (8), the indicator weight is determined. d b =1-e b , (10) Among them, z ab is the proportion of the ath evaluation object under the bth indicator, k is the number of evaluation individuals, e b is information entropy, d b is the information redundancy, w b is the indicator weight, w b ∈[0,1], 6. The evaluation method according to claim 5, characterized in that: The first evaluation value is obtained by using the TOPSIS method according to the determined weight, including: Establish a weighted decision matrix; Determining a positive ideal solution and a negative ideal solution according to the weighted decision matrix; Determining the distance between each evaluation object and the positive ideal solution and the negative ideal solution; Determine a ratio value between each evaluation solution and an ideal solution according to the distance; The evaluation schemes are ranked according to the ratio values ​​to determine a first evaluation value.

7. The evaluation method according to claim 6, characterized in that: Building a weighted decision matrix involves: According to formulas (12) and (13), the weighted decision matrix is ​​determined: R＝(r ab ) m×n ， (12) r ab =w b ×y ab , (13) Among them, R is the weighted decision matrix, r ab is the product of the indicator weight and the standardized value of the bth evaluation indicator of the ath evaluation object; Determining the positive ideal solution and the negative ideal solution according to the weighted decision matrix includes: According to formulas (14) and (15), the positive ideal solution and the negative ideal solution are determined. S + =max(r 1b ,r 2b ,…,r mb ), (14) S - =min(r 1b ,r 2b ,…,r mb ), (15) Among them, S + is a positive ideal solution, max(r 1b , r 2b ,…,r mb ) is the maximum value of each column in the weighted decision matrix, S - is a negative ideal solution, min(r 1b , r 2b ,…,r mb ) is the minimum value of each column in the weighted decision matrix; Determining the distance between each evaluation object and the positive ideal solution and the negative ideal solution includes: According to formulas (16) and (17), the distance between each evaluation object and the positive ideal solution and the negative ideal solution is determined. Among them, D + is the evaluation object and the positive ideal solution S + The distance between - Each evaluation object and the negative ideal solution S - The distance between a =(r a1 ,r a2 ,…,r ab ) is the vector of the ath row in the weighted decision matrix, S -1 is the inverse matrix of the covariance matrix; Determining the ratio of each evaluation solution to the ideal solution according to the distance includes: The ratio value is determined according to formula (18), Among them, C a is the proportional value.

8. The evaluation method according to claim 1, characterized in that: Based on the feature extracted data, obtaining the second evaluation value according to the SOFM method includes: According to formulas (19) and (20), cluster analysis is performed on the feature-extracted data. c=argmin α ||γ-in α ||, (19) Among them, c is the best matching neuron, γ is the input data, and w α is the weight vector of all neurons, argmin α ||γ-w α || is the input data γ and the weight vector w of all neurons α The neuron with the closest distance between them, α is the index of the neuron; w α (t+1)=w α (t)+θ(p(t)-w α (t-1)), (20) Among them, α∈N α* (d), N α* (d) is the best matching neuron α * is the center and all hidden neurons with radius d, θ is the learning rate, w α (t+1), w α (t), w α (t-1) represents the weight vector of the αth neuron at t+1, t, and t-1 time steps, p(t) is the output vector at the tth time step, and t is the time step.

9. The evaluation method according to claim 1, characterized in that: When it is determined that the first evaluation value and the second evaluation value are consistent, calculating the first comprehensive evaluation value of the current manufacturing enterprise includes: According to formula (21), the first comprehensive evaluation value is obtained: P T =0.5P1+0.5P2, (21) Among them, P T is the comprehensive evaluation value, P1 is the first evaluation value, and P2 is the second evaluation value.

10. An evaluation system for ranking manufacturing enterprises based on UISHSIC-entropy weight-TOPSIS and SAE-SOFM, characterized in that: The evaluation system comprises the evaluation method according to any one of claims 1 to 9.

Images