Material supervision data abnormal value identification method based on wavelet transform and Transform network

Abstract

The invention discloses a material supervision data abnormal value identification method based on wavelet transform and a Transform network, and relates to the technical field of material supervision data processing, and the method comprises the following steps: selecting target data, carrying out the normalization processing of the data, and obtaining the preprocessed data; performing feature extraction on the preprocessed data based on wavelet transform to obtain feature data; inputting the feature data into a transform network, carrying out clustering, and calculating Euclidean distances between all the data and corresponding clustering centers; and according to the calculated Euclidean distance, selecting a judgment standard to realize identification of abnormal data. According to the method, an advanced informatization technical means is adopted, abnormal values in mass material supervision data are screened and identified, a data basis is provided for subsequent digital-intelligent material purchase supervision by adopting an artificial intelligence technology, and a solid foundation and guarantee are provided for benign operation of enterprises.

Description

Technical Field

[0001] The present invention relates to the technical field of material supervision data processing. More specifically, the present invention relates to a method for identifying outliers in material supervision data based on wavelet transform and Transformer network. Background Art

[0002] In the operation and management of modern enterprises, material procurement is one of the essential links. The compliance and orderly development of material procurement are one of the prerequisites for ensuring the good operation of the enterprise. In material procurement, it is necessary to supervise the procurement process based on historical procurement-related data records to ensure the compliance and orderly development of material procurement. The effectiveness of material supervision is greatly affected by the supervision data. In actual work, due to the negligence of manual data recording, there are often some abnormal data, which seriously affects the effectiveness of material supervision data. The present invention adopts advanced information technology means to screen and identify outliers in material supervision data, providing a data basis for subsequent intelligent material procurement supervision using artificial intelligence technology, and providing a solid foundation and guarantee for the sound operation of the enterprise.

[0003] In the above-mentioned disclosed technical solution, there are at least the following technical problems: Due to the negligence of manual data recording, there are often some abnormal data, which seriously affects the effectiveness of material supervision data.

[0004] In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for identifying outliers in material supervision data based on wavelet transform and Transformer network. By adopting advanced information technology means, outliers in a large amount of material supervision data are screened and identified, providing a data basis for subsequent intelligent material procurement supervision using artificial intelligence technology, so as to solve the problem that due to the negligence of manual data recording, there are often some abnormal data, which seriously affects the effectiveness of material supervision data.

[0006] To achieve the above object, the present invention provides the following technical solution: A method for identifying outliers in material supervision data based on wavelet transform and Transformer network, comprising the following steps: selecting target data, performing normalization processing on the data to obtain preprocessed data; extracting features from the preprocessed data based on wavelet transform to obtain feature data; inputting the feature data into a transformer network, and performing clustering, calculating the Euclidean distance between all data and the corresponding cluster centers; according to the calculated Euclidean distance, selecting a judgment criterion to achieve the identification of abnormal data.

[0007] In a preferred embodiment, the selection of target data and the normalization of the data to obtain the preprocessed data are specifically as follows: Set the data to be processed, calculate the average value of the data to be processed, and perform normalization processing on the data to be processed based on the normalization formula to obtain the preprocessed data.

[0008] In a preferred embodiment, the feature extraction of the preprocessed data based on wavelet transform to obtain feature data is specifically as follows: Decompose the normalized data using wavelet transform to obtain wavelet basis functions; Perform an inner product operation on the basis functions and the original data to decompose the original data into sub-signals of different scales to obtain wavelet decomposition coefficients; Calculate the corresponding energy entropy according to the wavelet decomposition coefficients, and combine the energy entropies of all subspaces in sequence to obtain the final energy entropy feature set.

[0009] In a preferred embodiment, input the feature data into the transformer network and perform clustering, and calculate the Euclidean distance between all data and the corresponding cluster centers, specifically as follows: Use the feature set composed of the energy entropies of each subspace as the input of the encoder of the Transformer network model, use the clustering result and the Euclidean distance between each data and its cluster center as the input of the decoder, and further splice the output results of the decoder and the encoder after mapping through a fully connected layer to obtain the output corresponding one-to-one to the input data and the Euclidean distance.

[0010] In a preferred embodiment, the attention mechanism includes an attention function, and the attention function includes a scaled dot product that queries all K vectors to obtain the output matrix of the attention function.

[0011] In a preferred embodiment, according to the calculated Euclidean distance, select a judgment criterion to realize the identification of abnormal data, specifically as follows: Perform statistical analysis on the Euclidean distances between all calculated data points and the corresponding cluster centers; Draw a histogram of the Euclidean distances to observe the distribution pattern of the distances and judge whether it follows a normal distribution; If it follows a normal distribution, compare the Euclidean distance between each data point and the corresponding cluster center with the set judgment threshold; Determine the data points with Euclidean distances greater than the judgment threshold as abnormal data.

[0012] The technical effects and advantages of a method for identifying outliers in material supervision data based on wavelet transform and Transformer network of the present invention: 1. An outlier identification method for material supervision data based on wavelet transform and Transformer network of the present invention uses advanced information technology means to screen and identify outliers in a large amount of material supervision data, providing a data basis for subsequent digital material procurement supervision using artificial intelligence technology, and providing a solid foundation and guarantee for the sound operation of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic flowchart of an outlier identification method for material supervision data based on wavelet transform and Transformer network of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0015] Embodiment 1 Figure 1 An outlier identification method for material supervision data based on wavelet transform and Transformer network of the present invention is given, including the following steps: Select target data, perform normalization processing on the data to obtain the preprocessed data; Extract features from the preprocessed data based on wavelet transform to obtain feature data; Input the feature data into the transformer network, perform clustering, and calculate the Euclidean distance between all data and the corresponding cluster centers; According to the calculated Euclidean distance, select a judgment criterion to achieve the identification of abnormal data.

[0016] S1, select target data, perform normalization processing on the data to obtain the preprocessed data; The selection of target data and the normalization processing of the data to obtain the preprocessed data are specifically as follows: Set the data to be processed, calculate the average value of the data to be processed, and perform normalization processing on the data to be processed based on the normalization formula to obtain the preprocessed data.

[0017] The average value of the data to be processed is specifically as follows:

[0018] The normalization processing of the data to be processed based on the normalization formula is specifically as follows:

[0019] Among them, is the average value of the data to be processed, is the nth data, is the number of data to be processed, is the data after normalization processing, is the maximum value among all the data to be processed, is the minimum value among all the data to be processed.

[0020] Selecting the target data and performing normalization processing has the following advantages: Improving model accuracy: The material supervision data may have different dimensions and value ranges. Through normalization, the data is mapped to the same scale, avoiding the larger numerical range in the data from dominating the model training, enabling the model to treat each feature more fairly, and thus improving the accuracy of the model in identifying outliers.

[0021] Accelerating model convergence: The normalized data can make the gradient descent algorithm converge faster. When training the model, the scale differences in the data may lead to inconsistent step sizes of gradient updates, making the model training process unstable and even difficult to converge. Normalization processing can make the data have a similar change range in each dimension, helping the optimization algorithm find the optimal solution faster, reducing the training time and consumption of computing resources.

[0022] Enhancing data stability: Normalization can reduce the volatility of the data and make the data more stable. For some noise or abnormal fluctuations in the material supervision data, normalization can smooth these interference factors to a certain extent, improve the data quality, making it easier to identify outliers without being masked by the original fluctuations of the data.

[0023] Improving the model's generalization ability: The data after normalization processing can make the model have better generalization ability on different data sets. Because normalization converts the data into a general format, reducing the impact of data distribution differences on the model, enabling the model to better adapt to new and unseen data, and improving the ability to identify outliers in various material supervision data, rather than just performing well on the training data.

[0024] S2. Extract features from the preprocessed data based on wavelet transform to obtain feature data; The extracting features from the preprocessed data based on wavelet transform to obtain feature data is specifically as follows: Decompose the normalized data using wavelet transform to obtain wavelet basis functions; Perform an inner product operation on the basis functions and the original data, decompose the original data into sub-signals of different scales, and obtain wavelet decomposition coefficients; According to the wavelet decomposition coefficients, the corresponding energy entropy is calculated, and the energy entropies of all subspaces are combined in sequence to finally obtain the energy entropy feature set.

[0025] The wavelet basis function is specifically:

[0026] Among them, and respectively represent the basis functions obtained from the original data under different impulse functions, m is the wavelet decomposition scale, represents the position of the current data in all data sequences, and represent the impulse responses of the corresponding data; for the given data, the data can be decomposed into subspaces at the decomposition scale m.

[0027] The wavelet decomposition coefficients are specifically:

[0028] In the formula represents the wavelet decomposition coefficient obtained by decomposition, represents the data after normalization; represents the basis function value at the point (m, p); <> represents the inner product calculation.

[0029] The energy entropy feature set is specifically:

[0030] In the formula, E represents the total energy, represents the energy of the m-th subspace, Q represents the relative energy of the wavelet coefficients, and T represents the Shannon energy entropy of the m-th subspace.

[0031] S3. Input the feature data into the transformer network, perform clustering, and calculate the Euclidean distance between all data and the corresponding cluster centers; The step of inputting the feature data into the transformer network, performing clustering, and calculating the Euclidean distance between all data and the corresponding cluster centers is specifically: Use the feature set composed of the energy entropies of each subspace as the input of the encoder of the Transformer network model, use the clustering result and the Euclidean distance between each data and its cluster center as the input of the decoder, and further splice the output results of the decoder and the encoder after mapping through the fully connected layer to obtain the output in one-to-one correspondence with the input data and the Euclidean distance; The Transformer network also includes a single attention mechanism processing step, specifically:

[0032]

[0033]

[0034] Where R, K, V represent query vector, key vector and value vector respectively; , and They represent the coefficient matrices of the corresponding phasors in the self-attention mechanism, is the feature data vector.

[0035] The attention mechanism includes an attention function, and the attention function includes querying the scaled dot product of all K vectors to obtain an output matrix of the attention function; The output matrix is ​​specifically:

[0036] in, represents the output of the attention function, Represents the dimension of the input K vector.

[0037] The Euclidean distance is specifically:

[0038] in, represents the square of the Euclidean distance between vector x and vector y, n is the dimension of the vector, and are two n-dimensional vectors respectively.

[0039] S4, according to the calculated Euclidean distance, select the judgment standard to realize the identification of abnormal data.

[0040] According to the calculated Euclidean distance, a judgment standard is selected to realize the identification of abnormal data, specifically: Perform statistical analysis on the Euclidean distances between all the calculated data points and the corresponding cluster centers; Draw a histogram of the Euclidean distance, observe the distribution of the distance, and determine whether it obeys the normal distribution; If it obeys normal distribution, the Euclidean distance between each data point and the corresponding cluster center is compared with the set judgment threshold; The data points whose Euclidean distance is greater than the judgment threshold are judged as abnormal data.

[0041] The above formulas are all dimensionless and numerical calculations. The formula is a formula obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The technicians in the formula set it according to the actual situation.

[0042] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product.

[0043] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A professional technician can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0044] In addition, the functional modules in the various embodiments of the present application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0045] As described above, the above is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0046] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying outliers in material supervision data based on wavelet transform and Transformer network, characterized in that: The following steps are involved: Select target data, normalize the data, and obtain preprocessed data; Perform feature extraction on the preprocessed data based on wavelet transform to obtain feature data; Input the feature data into the transformer network, perform clustering, and calculate the Euclidean distance between all data and the corresponding cluster center; According to the calculated Euclidean distance, the judgment criteria are selected to realize the identification of abnormal data.

2. According to the method for identifying abnormal values ​​of material supervision data based on wavelet transform and Transformer network in claim 1, it is characterized in that: The target data is selected and normalized to obtain the pre-processed data, specifically: The data to be processed is set, the average value of the data to be processed is calculated, and the data to be processed is normalized based on the normalization formula to obtain the preprocessed data.

3. The method for identifying abnormal values ​​of material supervision data based on wavelet transform and Transformer network according to claim 2 is characterized in that: The pre-processed data is subjected to feature extraction based on wavelet transform to obtain feature data, specifically: Decompose the normalized data using wavelet transform to obtain wavelet basis functions; Perform inner product operation on basis function and original data, decompose original data into sub-signals of different scales, and obtain wavelet decomposition coefficients; According to the wavelet decomposition coefficients, the corresponding energy entropy is calculated, and the energy entropies of all subspaces are combined in sequence to obtain the final energy entropy feature set.

4. According to claim 3, a method for identifying abnormal values ​​of material supervision data based on wavelet transform and Transformer network is characterized in that: The feature data is input into the transformer network, clustered, and the Euclidean distance between all data and the corresponding cluster center is calculated, specifically: The feature set composed of the energy entropy of each subspace is used as the input of the Transformer network model encoder, and the clustering results and the Euclidean distance between each data and its cluster center are used as the input of the decoder. The output results of the decoder and the encoder are further mapped and concatenated through a fully connected layer to obtain an output with a one-to-one correspondence between the input data and the Euclidean distance.

5. According to claim 4, a method for identifying abnormal values ​​of material supervision data based on wavelet transform and Transformer network is characterized in that: According to the calculated Euclidean distance, a judgment standard is selected to realize the identification of abnormal data, specifically: Perform statistical analysis on the Euclidean distances between all the calculated data points and the corresponding cluster centers; Draw a histogram of the Euclidean distance, observe the distribution of the distance, and determine whether it obeys the normal distribution; If it obeys normal distribution, the Euclidean distance between each data point and the corresponding cluster center is compared with the set judgment threshold; The data points whose Euclidean distance is greater than the judgment threshold are judged as abnormal data.

6. According to claim 5, a method for identifying abnormal values ​​of material supervision data based on wavelet transform and Transformer network is characterized in that: The Transformer network also includes a single attention mechanism processing step, specifically: Where R, K, V represent query vector, key vector and value vector respectively; , and They represent the coefficient matrices of the corresponding phasors in the self-attention mechanism, is the feature data vector.

7. According to claim 6, a method for identifying abnormal values ​​of material supervision data based on wavelet transform and Transformer network is characterized in that: The attention mechanism includes an attention function, and the attention function includes querying the scaled dot products of all K vectors to obtain an output matrix of the attention function.

8. According to claim 7, a method for identifying abnormal values ​​of material supervision data based on wavelet transform and Transformer network is characterized in that: The energy entropy feature set is specifically: Where E represents the total energy, represents the energy of the m-th subspace, Q represents the relative energy of the wavelet coefficients, and T represents the Shannon energy entropy of the m-th subspace.

9. The method for identifying abnormal values ​​of material supervision data based on wavelet transform and Transformer network according to claim 8 is characterized in that: The wavelet basis function is specifically: in, and They represent the basis functions of the original data under different impulse functions, m is the wavelet decomposition scale, Indicates the position of the current data in all data sequences. and Represents the impulse response of the corresponding data; for a given data, the data can be decomposed into Subspace.

10. The method for identifying abnormal values ​​of material supervision data based on wavelet transform and Transformer network according to claim 9 is characterized in that: The output matrix is ​​specifically: in, represents the output of the attention function, Represents the dimension of the input K vector.

Images