Data anomaly detection method for cottonseed oil production

Abstract

The invention belongs to the technical field of cottonseed oil production, and particularly relates to a data anomaly detection method for cottonseed oil production. Firstly, sensor data of all links of raw material, oil pressing and refining are collected and preprocessed; then improving a model clustering algorithm, estimating clustering centers, introducing adaptive local weighted distance measurement to construct a new distance formula, iteratively calculating a membership matrix and an adaptive local weighted distance, substituting the membership matrix and the adaptive local weighted distance into a membership formula for dynamically adjusting an m value, and optimizing a clustering center set in combination with a simulated annealing thought; and judging the convergence of the membership matrix by using a Floribenius norm, and circulating until convergence. And finally, inputting real-time production data based on an improved algorithm to detect abnormity. The method accurately deals with complex working conditions of cottonseed oil production, improves data monitoring accuracy, effectively discriminates abnormal data, provides key technical support for guaranteeing production stability and improving product quality and equipment safety, assists enterprises in optimizing production management, and enhances market competitiveness.

Description

Technical Field

[0001] The invention belongs to the technical field of cottonseed oil production, and in particular relates to a data anomaly detection method used in cottonseed oil production. Background Art

[0002] The cottonseed oil production process covers complex links such as raw material pretreatment, oil pressing, and refining, involving the synergistic effect of multiple parameters. The accuracy of process parameters in each link is of far-reaching significance to product quality, production efficiency, and equipment life. With the upgrading of the industry and the advancement of intelligence, although production equipment has gradually realized automation, data monitoring and analysis technology lags behind. Traditional monitoring relies more on manual experience to judge the numerical values ​​of equipment instruments, which has strong subjective factors, large errors, and poor timeliness, making it difficult to meet the needs of modern efficient and precise production. Furthermore, the cottonseed oil production conditions are complex and changeable. Affected by raw material quality, equipment performance, and environmental factors, the data fluctuation pattern is difficult to follow, and weak anomalies are easy to accumulate and cause serious production accidents and quality defects. For example, abnormal pressure in the oil pressing link causes fluctuations in oil yield, and temperature inaccuracy in the refining stage affects the color and flavor of the oil. The existing general data monitoring method does not take into account the production characteristics of cottonseed oil. The data processing and analysis model has strong universality but weak pertinence. Under complex working conditions, it is easy to make frequent misjudgments or missed judgments, resulting in production management decision-making errors. Summary of the invention

[0003] In view of the technical problems existing in the above-mentioned background technology, the present invention proposes a data anomaly detection method for cottonseed oil production.

[0004] In order to achieve the above object, the technical solution adopted by the present invention comprises the following steps:

[0005] S1, first collect the data information of the sensor and pre-process the data;

[0006] S2, then improve the model clustering algorithm, and estimate the reasonable clustering center according to the normal working condition category of cottonseed oil production;

[0007] The improvement of the model clustering algorithm in step S2 is implemented as follows:

[0008] S21, introduce adaptive local weighted distance metric, construct adaptive local weighted distance in is the adaptive local weighted distance, n is the feature dimension, ω l is the adaptive weight of feature l, σ is the variance, c jl Represents the cluster center c j The value in the lth feature dimension, x il Represents data point x i The value on the lth feature dimension;

[0009] S22. In each iteration, the cluster center set obtained in the previous iteration is Where C is the number of clusters, and the membership matrix is ​​calculated based on the current data point set.

[0010] S23, for data point x i and cluster center c j , calculate the adaptive local weighted distance Substitute into the membership formula Where m is a fuzzy weighted index, which is adaptively adjusted according to the dynamic characteristics of production data;

[0011] S24, based on the current membership matrix and the data point set, the simulated annealing idea is introduced to optimize and calculate a new cluster center set;

[0012] S25, finally, compare the membership matrix before and after the iteration to determine whether it converges. If it converges, the iteration ends. If it does not converge, the cluster center and the membership matrix are updated until the convergence condition is met.

[0013] S3. Based on the clustering results obtained by the improved clustering algorithm, the real-time cottonseed production data is input to detect abnormal data.

[0014] Preferably, the sensor data information in step S1 includes temperature information, humidity information, and impurity detection information of the raw material pretreatment stage, chamber pressure information, screw speed information, and oil outlet temperature information of the oil pressing stage, deacidification tower temperature information, alkali solution flow information, and decolorization tank pressure information of the refining stage.

[0015] Preferably, the fuzzy weighted index m in step S23 is specifically implemented as follows:

[0016] S231, firstly setting the value of the initial fuzzy weighted index m;

[0017] S232. Calculate the distribution entropy E of the data points in each cluster. For the jth cluster, the entropy E j The calculation formula is: Distribution entropy of the entire dataset

[0018] S233, adjust m according to the value of entropy, Where E max , E min are the maximum and minimum entropy values ​​observed during the iterations, respectively.

[0019] Preferably, the specific implementation of introducing the simulated annealing idea to optimize and calculate the new cluster center set in step S24 is as follows:

[0020] S241, first set the initial temperature and cooling rate;

[0021] S242, calculate the objective function value As an energy function, it measures the quality of clustering;

[0022] S243, then generate a new set of candidate cluster centers Through C t The middle part of the cluster center is obtained by random perturbation in a small range;

[0023] S244. For each cluster center For each feature dimension l, generate a random number Δ that follows a normal distribution jl , then each feature dimension value of the candidate cluster center is

[0024] S245, then compare the objective function value, if J(C t* )<J(C t ) then accept the candidate cluster center, i.e. C t+1 =C t* , if J(C t* )≥J(C t ), then the candidate cluster center is accepted with probability p;

[0025] S246. Finally, the temperature is updated according to the cooling rate to obtain the final cluster center.

[0026] Preferably, the probability in step S245 is Where T t is the current temperature value of the simulated annealing algorithm.

[0027] Preferably, in step S25, the membership matrix before and after the iteration is compared, and the criterion for judging whether it converges is: t+1 -U t || F <0.05, where || || F is the Frobenius norm.

[0028] Compared with the prior art, the advantages and positive effects of the present invention are that the improved model-based clustering algorithm of the present invention has outstanding improvement advantages. The adaptive local weighted distance metric considers the multi-dimensional relationship and distribution characteristics between data points and cluster centers, dynamically adjusts the fuzzy weighted index, accurately reflects data differences, improves clustering accuracy, effectively handles complex production data fluctuations, and reduces misjudgments and missed judgments. And the simulated annealing optimization is introduced to help the algorithm escape from the local optimum. When cottonseed oil production is disturbed by raw materials, equipment, and environmental factors, resulting in complex and changeable data, the optimal cluster center is accurately located, and small anomalies are keenly captured to ensure stable production, constant quality, and equipment safety, thereby improving the level and benefits of intelligent management of the industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1 The present invention is implemented as a flow chart. DETAILED DESCRIPTION

[0031] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0032] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0033] Embodiment. In the field of cottonseed oil production, it is crucial to ensure the stability of the production process, consistent product quality, and safe operation of equipment. However, traditional data monitoring methods are unable to cope with the complex and changeable cottonseed oil production conditions. In the oil pressing process, if there is an abnormal fluctuation in the chamber pressure, it may cause a sudden drop in oil yield and a decline in oil quality. In the refining stage, the temperature of the deacidification tower or the flow of alkali liquid is inaccurate, which can easily lead to quality defects such as excessive acid value and darker color of the product. In order to accurately monitor abnormalities in cottonseed oil production data and ensure efficient and stable production, the present invention proposes a data anomaly detection method for cottonseed oil production. The implementation process of the present invention is as follows: Figure 1 shown.

[0034] In order to effectively collect key data in each link of cottonseed oil production and improve data quality, and solve the problems of misjudgment and missed judgment of abnormal monitoring caused by noise interference and data redundancy, a comprehensive preprocessing solution is adopted. The collected sensor data information includes temperature information, humidity information, and impurity detection information in the raw material pretreatment link, the pressing chamber pressure information, screw speed information, and oil outlet temperature information in the oil pressing link, and the deacidification tower temperature information, alkali liquid flow information, and decolorization tank pressure information in the refining stage. For each dimension of the collected data tensor, wavelet transform is first used for denoising. According to the frequency characteristics of the data, the adaptive wavelet basis function is selected, and the data is decomposed into multi-layer approximation and detail coefficients. The Stein unbiased risk threshold is used to process the detail coefficients for denoising and reconstruction. After PCA dimensionality reduction, the covariance matrix calculates the eigenvalues ​​and eigenvectors, and the principal components are selected according to the proportion of eigenvalues. The preprocessing matrix is ​​constructed with the linear combination coefficients of the data after dimensionality reduction, and the subsequent algorithm is enabled to process efficiently with low-dimensional high-quality data.

[0035] In order to optimize the clustering algorithm to accurately classify cottonseed oil production data and accurately locate abnormal data, the model clustering algorithm is deeply improved. Adaptive local weighted distance measurement and dynamic adjustment of m-value membership formula are introduced to fit the complex pattern of production data and improve clustering accuracy and stability. Simulated annealing is used to optimize cluster center calculation, break through local optimality, and enhance the sensitivity of anomaly detection. Adaptive local weighted distance measurement is introduced to construct adaptive local weighted distance in is the adaptive local weighted distance, n is the feature dimension, ω l is the adaptive weight of feature l, σ is the variance, c jl Represents the cluster center c j The value in the lth feature dimension, x il Represents data point x i The value on the lth feature dimension; in each iteration, the cluster center set obtained in the previous iteration Where C is the number of clusters, and the membership matrix is ​​calculated based on the current data point set. For a data point x i and cluster center c j , calculate the adaptive local weighted distance Substitute into the membership formula Where m is the fuzzy weighted index, which is adaptively adjusted according to the dynamic characteristics of production data; the specific implementation of the fuzzy weighted index m is as follows: first, the value of the initial fuzzy weighted index m is set; the distribution entropy E of the data points in each cluster is calculated. For the jth cluster, the entropy E j The calculation formula is: Distribution entropy of the entire dataset Adjust m according to the value of entropy, Where E max , E minThey are the maximum and minimum entropy values ​​observed during the iteration process. At the same time, based on the current membership matrix and data point set, the simulated annealing idea is introduced to optimize the calculation of the new cluster center set; first set the initial temperature and cooling rate; calculate the objective function value As an energy function, measure the quality of clustering; then generate a new set of candidate cluster centers Through C t The middle part of the cluster center is obtained by small-scale random perturbation; for each cluster center For each feature dimension l, generate a random number Δ that follows a normal distribution jl , then each feature dimension value of the candidate cluster center is Then compare the objective function values. If J(C t* )<J(C t ) then accept the candidate cluster center, i.e. C t+1 =C t* , if J(C t* )≥J(C t ), then the candidate cluster center is accepted with probability p; where the probability Where T t is the current temperature value of the simulated annealing algorithm, and the temperature is updated according to the cooling rate to obtain the final cluster center. Finally, the membership matrix before and after the iteration is compared to determine whether it converges. If it converges, the iteration ends. If it does not converge, the cluster center and the membership matrix are updated until the convergence condition is met; among them, the comparison of the membership matrix before and after the iteration to determine whether it converges is: ||U t+1 -Ut|| F <0.05, where || || F is the Frobenius norm. The deep improvement of the model clustering algorithm has multiple advantages. The introduction of adaptive local weighted distance measurement and dynamic adjustment of the m-value membership formula fits the complex pattern of production data and improves clustering accuracy and stability. Simulated annealing optimizes cluster center calculation, breaks through local optimality, enhances the sensitivity of anomaly detection, accurately captures minor anomalies, and effectively guarantees the stable operation of cottonseed oil production.

[0036] Finally, in order to accurately detect abnormal data in cottonseed oil production in real time and ensure the smooth and stable production process, based on the results of the improved clustering algorithm, the real-time cottonseed production data is input to detect abnormalities. In the continuous production process, new data is connected to the improved algorithm model and compared with the clustering results. It helps operators to respond quickly, stabilize production quality and efficiency, reduce failure losses, and improve the level of intelligent and precise management and control of the entire production process, providing solid technical support for corporate benefits and product competitiveness.

[0037] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for detecting data anomalies in cottonseed oil production, characterized in that: The following steps are involved: S1, first collect the data information of the sensor and pre-process the data; S2, then improve the model clustering algorithm, and estimate the reasonable clustering center according to the normal working condition category of cottonseed oil production; The improvement of the model clustering algorithm in step S2 is implemented as follows: S21, introduce adaptive local weighted distance metric, construct adaptive local weighted distance in is the adaptive local weighted distance, n is the feature dimension, ω l is the adaptive weight of feature l, σ is the variance, c jl Represents the cluster center c j The value in the lth feature dimension, x il Represents data point x i The value on the lth feature dimension; S22. In each iteration, the cluster center set obtained in the previous iteration is Where C is the number of clusters, and the membership matrix is ​​calculated based on the current data point set. S23, for data point x i and cluster center c j , calculate the adaptive local weighted distance Substitute into the membership formula Where m is a fuzzy weighted index, which is adaptively adjusted according to the dynamic characteristics of production data; S24, based on the current membership matrix and the data point set, the simulated annealing idea is introduced to optimize and calculate a new cluster center set; S25, finally, compare the membership matrix before and after the iteration to determine whether it converges. If it converges, the iteration ends. If it does not converge, the cluster center and the membership matrix are updated until the convergence condition is met. S3. Based on the clustering results obtained by the improved clustering algorithm, the real-time cottonseed production data is input to detect abnormal data.

2. The method for detecting data anomaly in cottonseed oil production according to claim 1, characterized in that: The sensor data information in step S1 includes temperature information, humidity information, and impurity detection information in the raw material pretreatment stage, chamber pressure information, screw speed information, and oil outlet temperature information in the oil pressing stage, and deacidification tower temperature information, alkali solution flow information, and decolorization tank pressure information in the refining stage.

3. The method for detecting data anomaly in cottonseed oil production according to claim 1, characterized in that: The specific implementation of the fuzzy weighted index m in step S23 is: S231, firstly setting the value of the initial fuzzy weighted index m; S232. Calculate the distribution entropy E of the data points in each cluster. For the jth cluster, the entropy E j The calculation formula is: Distribution entropy of the entire dataset S233, adjust m according to the value of entropy, Where E max , E min are the maximum and minimum entropy values ​​observed during the iterations, respectively.

4. The method for detecting data anomaly in cottonseed oil production according to claim 1, characterized in that: The specific implementation of introducing the simulated annealing idea to optimize and calculate the new cluster center set in step S24 is as follows: S241, first set the initial temperature and cooling rate; S242. Calculate the objective function value As an energy function, it measures the quality of clustering; S243, then generate a new set of candidate cluster centers Through C t The middle part of the cluster center is obtained by random perturbation in a small range; S244. For each cluster center For each feature dimension l, generate a random number Δ that follows a normal distribution jl , then each feature dimension value of the candidate cluster center is S245, then compare the objective function value, if Then accept the candidate cluster center, that is like Then accept the candidate cluster center with probability p; S246. Finally, the temperature is updated according to the cooling rate to obtain the final cluster center.

5. The method for detecting data anomaly in cottonseed oil production according to claim 4, characterized in that: The probability in step S245 Where T t is the current temperature value of the simulated annealing algorithm.

6. The method for detecting data anomaly in cottonseed oil production according to claim 1, characterized in that: In step S25, the membership matrix before and after iteration is compared to determine whether it converges: t+1 -U t || F <0.05, where |||| F is the Frobenius norm.

Images