Method and device for realizing main data archiving processing based on extreme learning machine under xinchuang environment, processor and readable storage medium thereof

Abstract

The application relates to a method for realizing main data archiving processing based on an extreme learning machine in a Xinxing environment, comprising the following steps: collecting and labeling relevant data of main data archiving; constructing an extreme learning machine classifier which fuses an ant colony optimization algorithm and a niche algorithm; inputting data samples to be archived, and using the trained classifier to predict the archiving priority of each data sample. The application also relates to a device, a processor and a storage medium for realizing main data archiving processing based on an extreme learning machine in a Xinxing environment. The method, the device, the processor and the computer readable storage medium for realizing main data archiving processing based on an extreme learning machine in a Xinxing environment can effectively collect and label relevant data of main data archiving of various business systems in an enterprise, make the data collection and labeling process clearer and more standardized, improve the performance and generalization ability of the model, improve the accuracy and stability of the main data archiving, and improve the convergence speed and performance of the model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computer software, especially to the field of data archiving, and in particular to a method and device for implementing main data archiving processing based on extreme learning machine in a Xiongguanneng environment, a processor and a computer readable storage medium thereof. BACKGROUND

[0002] In today's information age, enterprises are faced with a large amount of complex and diverse data, which usually comes from different business systems, channels and platforms. Among these data, main data plays a crucial role, representing the most important, most commonly used and most shared core data in the enterprise, such as customer information, product information, order information, etc. The accuracy, consistency and uniqueness of main data are crucial for the enterprise's business decision-making, business processes and customer service, etc.

[0003] Due to the importance of main data, enterprises usually need to store and manage these data in a unified and centralized manner, i.e. main data archiving. The purpose of main data archiving is to ensure the uniqueness, accuracy and completeness of data, so that the enterprise can quickly and accurately obtain the required information and support the decision-making and business operation of the enterprise.

[0004] However, in the Xiongguanneng environment, main data archiving faces a series of challenges. First, in the Xiongguanneng era, data types are diverse, covering unstructured data, large-scale data and high-dimensional data, etc. Traditional methods are difficult to effectively handle such complex and diverse data forms and characteristics, resulting in poor archiving results. Second, Xiongguanneng data changes rapidly, and traditional methods may not be able to adapt to the changing data environment, resulting in poor model generalization ability. In addition, the processing of large-scale data may also lead to low efficiency of traditional methods.

[0005] Therefore, in order to solve the problems faced by main data archiving in the Xiongguanneng environment, an innovative method needs to be proposed, which can adapt to complex and diverse Xiongguanneng data, improve archiving efficiency and accuracy, and has self-adaptive ability and individualized processing ability.

[0006] The existing technology still has the following problems:

[0007] 1. Insufficient data adaptability: Traditional main data archiving methods may not be able to adapt to complex and diverse data forms and data characteristics in the Xiongguanneng environment. Due to the diversity of data types and structures in the Xiongguanneng era, traditional methods may not be able to effectively handle unstructured data, large-scale data and high-dimensional data, etc., resulting in unsatisfactory archiving results.

[0008] 2. Low archiving efficiency: Traditional archiving methods may face the challenge of large data size and are inefficient in processing large amounts of data. The data collection, labeling, feature extraction and classification model training in the archiving process are time-consuming, affecting the real-time and efficiency of data archiving.

[0009] 3. Limited accuracy: Due to the diversity and complexity of data, traditional methods may not be able to fully exploit the potential rules and associations in the data, resulting in limited accuracy of the archiving results. Especially when facing highly variable signal creation data, the generalization ability of the model may be poor, resulting in reduced reliability of the archiving results. SUMMARY

[0010] The purpose of the present application is to overcome the above-mentioned shortcomings of the prior art, and to provide a method, device, processor and computer readable storage medium for realizing main data archiving processing based on extreme learning machine in a signal creation environment, which has good accuracy, simple operation and wide application range.

[0011] In order to achieve the above-mentioned purpose, the method, device, processor and computer readable storage medium for realizing main data archiving processing based on extreme learning machine in a signal creation environment according to the present application are as follows:

[0012] The main feature of the method for realizing main data archiving processing based on extreme learning machine in a signal creation environment is that the method comprises the following steps:

[0013] (1) Collect and label the relevant data of main data archiving;

[0014] (2) Construct an extreme learning machine classifier that combines ant colony optimization algorithm and niche algorithm;

[0015] (3) Input the data samples to be archived, and use the trained classifier to predict the archiving priority of each data sample.

[0016] Preferably, the data in step (1) is stored in table form, with each row representing a data sample and each column representing a feature or attribute.

[0017] Preferably, step (2) specifically comprises the following steps:

[0018] (2.1) Initialize the hidden layer weights and biases of the extreme learning machine;

[0019] (2.2) Use the ant colony optimization algorithm to select the number of hidden layer neurons;

[0020] (2.3) Use the niche algorithm for multi-objective optimization to optimize the output layer weights of the extreme learning machine;

[0021] (2.4) Perform adaptive learning rate adjustment based on cosine annealing;

[0022] (2.5) Use the optimized Extreme Learning Machine for text classification.

[0023] Preferably, step (2.2) specifically includes:

[0024] In each iteration, each ant selects a new number L′ of hidden layer neurons based on pheromone intensity and heuristic information. i And calculate the classification performance P of the extreme learning machine. i According to classification performance P i Update the pheromone concentration τ.

[0025] Preferably, updating the pheromone concentration τ in step (2.2) specifically involves:

[0026] Update the pheromone concentration τ according to the following formula:

[0027]

[0028] Where ρ is the pheromone volatility coefficient, δ(L′) i ) is the Kronecker function, P i For the classification performance of the Extreme Learning Machine, N a This refers to the size of the ant colony.

[0029] Preferably, step (2.3) specifically includes the following steps:

[0030] (2.3.1) In each iteration, each individual searches for new output layer weights β′ in the solution space based on its fitness. i And calculate the classification performance P′ of the extreme learning machine. i ;

[0031] (2.3.2) Select the output layer weights with the highest fitness as the output layer weights of the extreme learning machine.

[0032] Preferably, step (2.4) specifically includes the following steps:

[0033] Before training begins, set a lower bound η for the learning rate. min , upper limit η max and maximum number of iterations T max ;

[0034] At the beginning of each iteration, calculate the learning rate η for the t-th iteration. t ;

[0035] Each time the weights are updated, the current learning rate η is used. t Calculate the weights for the t-th iteration.

[0036] Preferably, in step (2.4), the learning rate η for the t-th iteration is calculated. t Specifically:

[0037] The learning rate η for the t-th iteration is calculated using the following formula. t :

[0038]

[0039] Where, η t η is the learning rate in the t-th iteration. min and η max These are the lower and upper bounds of the learning rate, T. cur It is the current iteration number, T max This is the maximum number of iterations set.

[0040] The calculation of the weights in step (2.4) for the t-th iteration is as follows:

[0041] The weights for the t-th iteration are calculated using the following formula:

[0042]

[0043] Where, β t It is the weight of the t-th iteration. The loss function L with respect to β t The gradient.

[0044] Preferably, step (2.5) specifically includes the following steps:

[0045] (2.5.1) Input the feature matrix X and calculate the hidden layer output H;

[0046] (2.5.2) Perform batch normalization on the hidden layer output H;

[0047] (2.5.3) Calculate the attention score s i ;

[0048] (2.5.4) Calculate attention weights Through attention weight a i Adjust the normalized hidden layer output Hs to obtain the new hidden layer output H′;

[0049] (2.5.5) Introduce dropout into the extreme learning machine and compute the updated hidden layer output H′;

[0050] (2.5.6) Substitute H″ into the output layer calculation formula of the extreme learning machine to obtain the output layer output O.

[0051] Preferably, the calculation of the hidden layer output H in step (2.5.1) is specifically as follows:

[0052] The hidden layer output H is calculated according to the following formula:

[0053] H = σ (XW + b) ;

[0054] where σ (·) is an activation function, X is an input feature matrix, W and b are a weight matrix and a bias vector of the hidden layer, respectively;

[0055] The step (2.5.2) of performing batch normalization on the hidden layer output H is specifically:

[0056] The hidden layer output H is batch-normalized according to the following formula:

[0057]

[0058] where Hs is the normalized hidden layer output, μ H is the mean of H, is the variance of H, and ∈ is a constant;

[0059] The step (2.5.3) of calculating the attention score s i is specifically:

[0060] The attention score s i is calculated according to the following formula:

[0061]

[0062] where w a is an attention weight, b a is a bias, h i is the i-th column of H, and tanh is an activation function;

[0063] The step (2.5.4) of calculating the attention weight w is specifically:

[0064] The attention weight w is calculated according to the following formula:

[0065]

[0066] The step (2.5.5) of calculating the updated hidden layer output H' is specifically:

[0067] The updated hidden layer output H' is calculated according to the following formula:

[0068] H'' = D ⊙ H';

[0069] where ⊙ represents Hadamard product, and D is a corresponding dropout mask;

[0070] The step (2.5.6) calculates the output layer output O, specifically:

[0071] The output layer output O is calculated according to the following formula:

[0072] O = H"β;

[0073] Wherein, β is the output layer weight, O is the classification result of the extreme learning machine.

[0074] The device for implementing the main data archiving processing based on the extreme learning machine in the XG environment, its main feature is, the device includes:

[0075] The processor is configured to execute computer executable instructions;

[0076] The memory stores one or more computer executable instructions, and the computer executable instructions are executed by the processor to realize each step of the method for implementing the main data archiving processing based on the extreme learning machine in the XG environment.

[0077] The processor for implementing the main data archiving processing based on the extreme learning machine in the XG environment, its main feature is, the processor is configured to execute computer executable instructions, and the computer executable instructions are executed by the processor to realize each step of the method for implementing the main data archiving processing based on the extreme learning machine in the XG environment.

[0078] The computer readable storage medium, its main feature is, it is stored with computer program, the computer program can be executed by the processor to realize each step of the method for implementing the main data archiving processing based on the extreme learning machine in the XG environment.

[0079] The method, device, processor and computer readable storage medium for realizing main data archiving processing based on an extreme learning machine in a Xireng environment of the application can effectively collect and label the main data archiving related data of various business systems in an enterprise. The data is stored in a table form and detailed data attributes and characteristics are described, so that the data collection and labeling process is clearer and more standardized. The extreme learning machine classifier combining the ant colony optimization algorithm and the niche algorithm is introduced to realize the classification prediction of the main data archiving task. The input data and output layer weight of the extreme learning machine are optimized through clustering sampling and multi-objective optimization, so that the performance and generalization ability of the model are improved. The attention mechanism and dropout technology are introduced into the extreme learning machine classifier to enhance the attention ability of the model to key features and improve the robustness of the model. The introduction of attention weights and dropout probabilities makes the model more suitable for different sample characteristics, and improves the accuracy and stability of the main data archiving. Through the cosine annealing adaptive learning rate adjustment strategy, the model can use a higher learning rate for fast learning in the early stage of model training, and gradually reduce the learning rate in the later stage to obtain more refined optimization effect, so that the convergence speed and performance of the model are improved. BRIEF DESCRIPTION OF DRAWINGS

[0080] Figure 1 The flowchart of the method for realizing main data archiving processing based on an extreme learning machine in a Xireng environment of the application. DETAILED DESCRIPTION

[0081] In order to more clearly describe the technical content of the application, the following further describes in combination with specific embodiments.

[0082] The method for realizing main data archiving processing based on an extreme learning machine in a Xireng environment of the application includes the following steps:

[0083] (1) collecting and labeling the related data of main data archiving;

[0084] (2) constructing an extreme learning machine classifier combining the ant colony optimization algorithm and the niche algorithm;

[0085] (3) inputting the data samples to be archived, and using the trained classifier to predict the archiving priority of each data sample.

[0086] As a preferred embodiment of the application, the data in step (1) is stored in a table form, and each row represents a data sample and each column represents a feature or attribute.

[0087] As a preferred embodiment of the application, step (2) specifically includes the following steps:

[0088] (2.1) Initialize the hidden layer weights and biases of the Extreme Learning Machine;

[0089] (2.2) Use the ant colony optimization algorithm to select the number of neurons in the hidden layer;

[0090] (2.3) Use the niche algorithm for multi-objective optimization to optimize the output layer weights of the extreme learning machine;

[0091] (2.4) Adaptive learning rate adjustment based on cosine annealing;

[0092] (2.5) Use the optimized Extreme Learning Machine for text classification.

[0093] In a preferred embodiment of the present invention, step (2.2) specifically comprises:

[0094] In each iteration, each ant selects a new number L′ of hidden layer neurons based on pheromone intensity and heuristic information. i And calculate the classification performance P of the extreme learning machine. i According to classification performance P i Update the pheromone concentration τ.

[0095] In a preferred embodiment of the present invention, updating the pheromone concentration τ in step (2.2) specifically involves:

[0096] Update the pheromone concentration τ according to the following formula:

[0097]

[0098] Where ρ is the pheromone volatility coefficient, δ(L′) i ) is the Kronecker function, P i For the classification performance of the Extreme Learning Machine, N a This refers to the size of the ant colony.

[0099] In a preferred embodiment of the present invention, step (2.3) specifically includes the following steps:

[0100] (2.3.1) In each iteration, each individual searches for new output layer weights β in the solution space based on its fitness. i And calculate the classification performance P′ of the extreme learning machine. i ;

[0101] (2.3.2) Select the output layer weights with the highest fitness as the output layer weights of the extreme learning machine.

[0102] In a preferred embodiment of the present invention, step (2.4) specifically includes the following steps:

[0103] Before training begins, set a lower bound η for the learning rate.min , upper limit η max , and maximum iteration number T max ;

[0104] At the beginning of each iteration, the learning rate η t of the t-th iteration is calculated.

[0105] At each weight update, the current learning rate η t is used to calculate the weight of the t-th iteration.

[0106] As a preferred embodiment of the present application, the step (2.4) of calculating the learning rate η t of the t-th iteration specifically includes the following steps:

[0107] The learning rate η t of the t-th iteration is calculated according to the following formula:

[0108]

[0109] where η t is the learning rate of the t-th iteration, η min and η max are the lower and upper limits of the learning rate, respectively, T cur is the current iteration number, and T max is the maximum iteration number set.

[0110] The step (2.4) of calculating the weight of the t-th iteration specifically includes the following steps:

[0111] The weight of the t-th iteration is calculated according to the following formula:

[0112]

[0113] where β t is the weight of the t-th iteration, and is the gradient of the loss function L with respect to β t .

[0114] As a preferred embodiment of the present application, the step (2.5) specifically includes the following steps:

[0115] (2.5.1) Input the feature matrix X and calculate the hidden layer output H;

[0116] (2.5.2) Perform batch normalization on the hidden layer output H;

[0117] (2.5.3) Calculate the attention score s i ;

[0118] (2.5.4) Calculate the attention weight a through the attention weight ai adjusting the normalized hidden layer output Hs to obtain a new hidden layer output H';

[0119] (2.5.5) introducing dropout in the extreme learning machine to calculate an updated hidden layer output H';

[0120] (2.5.6) substituting H" into the output layer calculation formula of the extreme learning machine to obtain an output layer output O.

[0121] As a preferred embodiment of the present application, the step (2.5.1) of calculating the hidden layer output H is specifically:

[0122] The hidden layer output H is calculated according to the following formula:

[0123] H = σ (XW + b) ;

[0124] wherein σ (·) is an activation function, X is an input feature matrix, W and b are a weight matrix and a bias vector of the hidden layer respectively;

[0125] The step (2.5.2) of performing batch normalization on the hidden layer output H is specifically:

[0126] The hidden layer output H is batch-normalized according to the following formula:

[0127]

[0128] wherein Hs is the normalized hidden layer output, μ H is the mean of H, is the variance of H, and ∈ is a constant;

[0129] The step (2.5.3) of calculating the attention score s i is specifically:

[0130] The attention score s i is calculated according to the following formula:

[0131]

[0132] wherein w a is an attention weight, b a is a bias, h i is the i-th column of H, and tanh is an activation function;

[0133] The step (2.5.4) of calculating the attention weight is specifically:

[0134] The attention weight is calculated according to the following formula:

[0135]

[0136] The step (2.5.5) calculates the updated hidden layer output H', specifically:

[0137] The updated hidden layer output H' is calculated according to the following formula:

[0138] H'' = D ▥ H';

[0139] Where ▥ represents Hadamard product, and D is the corresponding dropout mask.

[0140] The step (2.5.6) calculates the output layer output O, specifically:

[0141] The output layer output O is calculated according to the following formula:

[0142] O = H'' ▥ β;

[0143] Where β is the output layer weight, and O is the classification result of the extreme learning machine.

[0144] The device for implementing the main data archiving processing based on the extreme learning machine in the signal creation environment of the application, wherein the device comprises:

[0145] A processor configured to execute computer executable instructions;

[0146] A memory storing one or more computer executable instructions, wherein the computer executable instructions are executed by the processor to implement each step of the method for implementing the main data archiving processing based on the extreme learning machine in the signal creation environment.

[0147] The processor for implementing the main data archiving processing based on the extreme learning machine in the signal creation environment of the application, wherein the processor is configured to execute computer executable instructions, and the computer executable instructions are executed by the processor to implement each step of the method for implementing the main data archiving processing based on the extreme learning machine in the signal creation environment.

[0148] The computer readable storage medium of the application, wherein the computer program is stored on the computer readable storage medium, and the computer program can be executed by the processor to implement each step of the method for implementing the main data archiving processing based on the extreme learning machine in the signal creation environment.

[0149] The application proposes a new archiving method based on the extreme learning machine for the main data archiving task in the signal creation environment. The main data archiving refers to storing and managing the important, commonly used, and shared core data in a unified and centralized manner to ensure the uniqueness and accuracy of the data.

[0150] The application is a main data archiving method based on extreme learning machine in a signal creation environment, aiming to solve the problems and deficiencies of existing technologies in main data archiving tasks in a signal creation environment. The core of the application is to propose an extreme learning machine classifier that combines ant colony optimization algorithm and niche algorithm, and an optimization strategy that introduces attention mechanism and dropout technology.

[0151] Firstly, the application collects and annotates data related to main data archiving from various business systems within the enterprise through the data collection and annotation phase. These data are stored in table form, with each row representing a data sample and each column representing a feature or attribute. The data format and attributes are described in detail and demonstrated through specific examples.

[0152] Secondly, the key technology of the application is to build an extreme learning machine classifier that combines ant colony optimization algorithm and niche algorithm. The input training data of the extreme learning machine is optimized through clustering sampling, then the number of hidden layer neurons is selected using the ant colony optimization algorithm, and the output layer weights are optimized using the niche algorithm. Such a classifier can effectively improve the performance and generalization ability of the model, and better adapt to the diversified data characteristics in the signal creation environment.

[0153] Further, the application introduces attention mechanism and dropout technology to improve the attention ability and robustness of the model. Attention mechanism is used to adjust the hidden layer output, while dropout technology makes the model more robust and generalizable, improving the accuracy and stability of main data archiving.

[0154] Finally, the application optimizes the learning rate of the model through the cosine annealing adaptive learning rate adjustment strategy to improve the training efficiency and convergence speed. Such a strategy can make the model use a higher learning rate for fast learning in the early stage, and gradually reduce the learning rate in the later stage to obtain more refined optimization results, ultimately realizing the automatic archiving of main data.

[0155] In the specific embodiment of the application, a main data archiving method based on extreme learning machine in a signal creation environment is proposed, and the main steps are as follows:

[0156] Step 1: Data collection and annotation

[0157] Collect and annotate data related to main data archiving. Main data archiving refers to storing and managing important, commonly used, and shared core data in a unified and centralized manner to ensure data uniqueness and accuracy. The data comes from various business systems within the enterprise. The data is stored in table form, with each row representing a data sample and each column representing a feature or attribute. The data format is as follows:

[0158] Let X be the input data matrix, where m is the number of data samples and n is the number of features for each data sample.

[0159] Data attributes include core data from various business systems within the enterprise, such as customer information, product information, and order information. These data characteristics may include types such as text, numerical values, and dates.

[0160] Suppose you have a customer information table containing the following fields: Customer ID, Age, Location, Number of Purchases, and Total Spending. You want to use this data to train a master data archiving model to predict customer archiving priorities (e.g., high, medium, low) based on their information.

[0161] In this example, each row of data is treated as a sample, and each field as a feature. The data is represented using the following formula:

[0162] Let customer ID be x1, age be x2, location be x3, number of purchases be x4, and total spending be x5. The data matrix X is shown below:

[0163]

[0164] Where, x ij Let j represent the j-th feature of the i-th sample.

[0165] Step 2: Construct a classification model:

[0166] This invention proposes an extreme learning machine classifier that integrates ant colony optimization algorithm and niche algorithm.

[0167] First, a clustering-based sampling strategy is used to optimize the input training data for the Extreme Learning Machine.

[0168] Let T be the dataset input to the Extreme Learning Machine, where m are the number of samples and n are the number of features. The k-means algorithm is used to cluster T. The clustering result can be expressed as:

[0169] S = k-means(T,K)

[0170] Where S = {S1, S2, ..., S} K} represents the clustering result, and K is the number of clusters.

[0171] Then from each cluster S k If p samples are sampled from the sample, and the sampling proportion is α, then p = α|S k |, where |S k | is clustering S k The sample size. The sampling result is T′. k =sample(S k ,p), where T′ k From cluster S kThe sample set obtained by sampling is combined, and an optimized extreme learning machine training set T' is obtained:

[0172]

[0173] Wherein, S k is the kth cluster, sample(S k , p) is sampling p samples from the cluster S k .

[0174] Then, the extreme learning machine is trained by using T' instead of T.

[0175] Further, the extreme learning machine is optimized by fusing an ant colony optimization algorithm and a niche algorithm.

[0176] The ant colony optimization algorithm is a meta-heuristic optimization algorithm simulating the foraging behavior of ant colonies in nature. The algorithm is used to select the number of hidden layer neurons of the extreme learning machine.

[0177] The number of hidden layer neurons of the extreme learning machine is defined as the solution space of the problem, each ant selects a path in the solution space according to the intensity of pheromone and heuristic information, and each path corresponds to a number of hidden layer neurons. The ants search in the solution space, and the concentration of pheromone will be updated according to the feedback of the classification performance to guide the search.

[0178] The niche algorithm is a meta-heuristic algorithm simulating the search for the most suitable living space of biological individuals in the living environment. The algorithm is used to optimize the output layer weight of the extreme learning machine.

[0179] The output layer weight of the extreme learning machine is defined as the solution space of the problem, and each individual corresponds to a set of output layer weights. Each individual searches in the solution space according to its fitness, and the fitness is the classification performance of the extreme learning machine. The most suitable living space corresponds to the optimal output layer weight.

[0180] Next, the construction method of the extreme learning machine classifier algorithm fusing the ant colony optimization algorithm and the niche algorithm is described in detail.

[0181] 1. Initialize the hidden layer weight and bias of the extreme learning machine.

[0182] Suppose that the feature matrix obtained after feature extraction of the text is X, where m is the sample number and n is the feature number. Suppose that the number of hidden layer neurons is L, then the hidden layer weight matrix W and the bias vector b can be randomly initialized as follows:

[0183] W = rand(n, L); b = rand(L);

[0184] 2. Select the number of hidden layer neurons using the ant colony optimization algorithm.

[0185] Assume the size of the ant colony is N a , then each ant a i corresponds to a number of hidden layer neurons L i , L i is initialized to a random value. At each iteration, each ant selects a new number of hidden layer neurons L' according to the pheromone intensity and heuristic information i , and calculates the classification performance P of the extreme learning machine i . The concentration of pheromone τ is updated according to P i :

[0186]

[0187] where ρ is the evaporation coefficient of pheromone, δ(L' - L) is the Kronecker function, δ(L' - L) = 1 when L' is the current optimal solution, otherwise it is 0. i i i

[0188] 3. Use the niche algorithm for multi-objective optimization to optimize the output layer weights of the extreme learning machine.

[0189] Assume the size of the population is N p , then each individual p i corresponds to a set of output layer weights β i , β i is initialized to a random value. At each iteration, each individual searches for a new set of output layer weights β' in the solution space according to its fitness, and calculates the classification performance P' of the extreme learning machine i . The fitness f of the individual is updated according to P' i : i i

[0190] f = P' i i

[0191] Finally, the output layer weights with the highest fitness are selected as the output layer weights of the extreme learning machine.

[0192] 4. Based on cosine annealing adaptive learning rate adjustment.

[0193] Through the process of simulated annealing, this strategy can automatically use a higher learning rate in the early stage for rapid learning, and gradually reduce the learning rate in the later stage to obtain more refined optimization results.

[0194] Before training begins, set the lower limit of the learning rate η min , the upper limit η max , and the maximum number of iterations T max . ​​​​​​

[0195] At the beginning of each iteration, the current learning rate η is calculated using the following formula t The learning rate adjustment strategy of cosine annealing is as follows:

[0196]

[0197] where η t is the learning rate of the t-th iteration, η min and η max are the lower and upper limits of the learning rate, respectively, T cur is the current iteration number, and T max is the maximum number of iterations set.

[0198] At each weight update, the current learning rate η t is used instead of the fixed learning rate. For the weight update formula of the extreme learning machine, η t is used to replace the original learning rate:

[0199]

[0200] where β t is the weight of the t-th iteration, is the gradient of the loss function L with respect to β t .

[0201] 5. Use the optimized extreme learning machine for text classification.

[0202] The present application introduces an attention mechanism, defines an attention weight w a , and w a is composed of The number of hidden layer neurons is L. a i is the attention weight of the i-th hidden layer neuron.

[0203] Input the feature matrix X and calculate the hidden layer output H:

[0204] H = σ(XW + b)

[0205] where σ(·) is the activation function, X is the input feature matrix, W and b are the weight matrix and bias vector of the hidden layer, respectively.

[0206] Further, the hidden layer output H is batch normalized:

[0207]

[0208] where Hs is the normalized hidden layer output, μ H is the mean of H, is the variance of H, and ∈ is a very small number to prevent division by zero.

[0209] Further, the attention score s is calculated i :

[0210]

[0211] where w is the attention weight, b is the bias, h is the i-th column of H. tanh is the activation function. a a i

[0212] Further, the attention weight a is calculated

[0213]

[0214] The normalized hidden layer output Hs is adjusted with the attention weight a to obtain a new hidden layer output H': i

[0215] H' = Hs ⊙ a

[0216] where ⊙ is the Hadamard product.

[0217] Further, dropout is introduced in the extreme learning machine to make the model more robust, enhance its resistance to noise and outliers, and thus improve its generalization ability.

[0218] Let the dropout probability of a neuron be p, and the corresponding dropout mask be D, then the updated hidden layer output H" is:

[0219] H" = D ⊙ H'

[0220] where ⊙ represents the Hadamard product, i.e. element-wise multiplication.

[0221] Substitute H" into the output layer calculation formula of the extreme learning machine, and the output layer output O after applying dropout can be obtained:

[0222] O = H"β

[0223] where β is the output layer weight, and O is the classification result of the extreme learning machine.

[0224] Step three: main data archiving:

[0225] The above model is trained using training samples. After the model training is completed, input the data samples to be archived, and use the trained classifier to predict the archiving priority of each data sample.

[0226] The specific implementation scheme of the present embodiment can be referred to the related description in the above embodiments, which will not be repeated here.

[0227] ​​​​It can be understood that the same or similar parts in the above embodiments can be mutually referenced, and the content not described in detail in some embodiments can refer to the same or similar content in other embodiments.

[0228] It should be noted that, in the description of the present application, the terms "first", "second" and the like are used only for descriptive purposes, and cannot be construed as indicating or implying relative importance. In addition, in the description of the present application, unless otherwise specified, the meaning of "a plurality of" is at least two.

[0229] Any process or method descriptions in flow charts or described herein otherwise can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for performing specific logic functions or steps in the process, and the various embodiments of the present application include additional implementations in which the functions described with reference to a given code module, segment, or portion of code are combined with functions described with reference to one or more other code modules, segments, or portions of code. The order in which a process is described is not necessarily the order in which the processes can be performed.

[0230] It should be understood that each part of the present application can be realized by hardware, software, firmware or their combination. In the above embodiments, a plurality of steps or methods can be realized by software or firmware stored in a memory and executed by a suitable instruction execution device. For example, if realized by hardware, and as in another embodiment, it can be realized by any one or their combination of the following technologies known in the art: discrete logic circuit with logic gate circuit for implementing logic function on data signal, application specific integrated circuit with suitable combination logic gate circuit, programmable gate array (PGA), field programmable gate array (FPGA) and the like.

[0231] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiment method can be completed by a program instructing the relevant hardware, and the corresponding program can be stored in a computer readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0232] In addition, each functional unit in each embodiment of the present application can be integrated in one processing module, or each unit can exist physically independently, or two or more units can be integrated in one module. The above integrated module can be realized in the form of hardware or in the form of software function module. The integrated module, if realized in the form of software function module and sold or used as an independent product, can also be stored in a computer readable storage medium.

[0233] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0234] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in one or more embodiments or examples.

[0235] The method, device, processor and computer readable storage medium for realizing main data archiving processing based on extreme learning machine in the Xiongxin environment of the application can effectively collect and label the main data archiving related data of each business system in the enterprise. The data is stored in the form of a table, and detailed data attributes and characteristics are described, so that the data collection and labeling process is more clear and standardized. The extreme learning machine classifier combining the ant colony optimization algorithm and the niche algorithm is introduced in the application, which is used to realize the classification prediction of the main data archiving task. The input data and output layer weight of the extreme learning machine are optimized through clustering sampling and multi-objective optimization, which improves the performance and generalization ability of the model. The attention mechanism and dropout technology are introduced in the extreme learning machine classifier to enhance the attention ability of the model to key features and improve the robustness of the model. The introduction of attention weight and dropout probability makes the model more suitable for the characteristics of different samples, improves the accuracy and stability of the main data archiving. Through the cosine annealing adaptive learning rate adjustment strategy, the application can use a higher learning rate for fast learning in the early stage of model training, and gradually reduce the learning rate in the later stage to obtain more fine optimization effect, which improves the convergence speed and performance of the model.

[0236] In this specification, the application has been described with reference to its specific embodiments. However, it is obvious that various modifications and changes can be made without departing from the spirit and scope of the application. Therefore, the specification and drawings should be considered illustrative rather than restrictive.

Claims

1. A method for master data archiving processing based on Extreme Learning Machine in a domestically developed information technology environment, characterized in that, The method includes the following steps: (1) Collect and label relevant data for master data archiving; (2) Construct an extreme learning machine classifier that integrates ant colony optimization algorithm and niche algorithm; (3) Input the data samples to be archived and use the trained classifier to predict the archiving priority of each data sample; Step (2) specifically includes the following steps: (2.1) Initialize the hidden layer weights and biases of the Extreme Learning Machine; (2.2) Use the ant colony optimization algorithm to select the number of neurons in the hidden layer; (2.3) Use the niche algorithm for multi-objective optimization to optimize the output layer weights of the extreme learning machine; (2.4) Adaptive learning rate adjustment based on cosine annealing; (2.5) Use the optimized Extreme Learning Machine for text classification.

2. The method for master data archiving processing based on Extreme Learning Machine in a domestically developed information technology environment according to claim 1, characterized in that, The data in step (1) is stored in tabular form, with each row representing a data sample and each column representing a feature or attribute.

3. The method for master data archiving processing based on Extreme Learning Machine in a domestically developed information technology environment according to claim 1, characterized in that, The specific steps (2.2) are as follows: In each iteration, each ant selects a new number L′ of hidden layer neurons based on pheromone intensity and heuristic information. i And calculate the classification performance P of the extreme learning machine. i According to classification performance P i Update the pheromone concentration τ.

4. The method for master data archiving processing based on Extreme Learning Machine in a domestically developed information technology environment, as described in claim 3, is characterized in that... In step (2.2), updating the pheromone concentration τ specifically involves: Update the pheromone concentration τ according to the following formula: Where ρ is the pheromone volatility coefficient, δ(L′) i ) is the Kronecker function, P i For the classification performance of the Extreme Learning Machine, N a This refers to the size of the ant colony.

5. The method for master data archiving processing based on Extreme Learning Machine in a domestically developed information technology environment according to claim 1, characterized in that, Step (2.3) specifically includes the following steps: (2.3.1) In each iteration, each individual searches for new output layer weights β′ in the solution space based on its fitness. i And calculate the classification performance P of the extreme learning machine. i ′; (2.3.2) Select the output layer weights with the highest fitness as the output layer weights of the extreme learning machine.

6. The method for master data archiving processing based on Extreme Learning Machine in a domestically developed information technology environment according to claim 3, characterized in that, Step (2.4) specifically includes the following steps: Before training begins, set a lower bound η for the learning rate. min , upper limit η max and maximum number of iterations T max ; At the beginning of each iteration, calculate the learning rate η for the t-th iteration. t ; Each time the weights are updated, the current learning rate η is used. t Calculate the weights for the t-th iteration.

7. The method for master data archiving processing based on Extreme Learning Machine in a domestically developed information technology environment according to claim 6, characterized in that, In step (2.4), the learning rate η for the t-th iteration is calculated. t Specifically: The learning rate η for the t-th iteration is calculated using the following formula. t : Where, η t η is the learning rate in the t-th iteration. min and η max These are the lower and upper bounds of the learning rate, T. cur It is the current iteration number, T max This is the maximum number of iterations set. The calculation of the weights in step (2.4) for the t-th iteration is as follows: The weights for the t-th iteration are calculated using the following formula: Where, β t It is the weight of the t-th iteration. The loss function L with respect to β t The gradient.

8. The method for master data archiving processing based on Extreme Learning Machine in a domestically developed information technology environment according to claim 1, characterized in that, Step (2.5) specifically includes the following steps: (2.5.1) Input the feature matrix X and calculate the hidden layer output H; (2.5.2) Perform batch normalization on the hidden layer output H; (2.5.3) Calculate the attention score s i ; (2.5.4) Calculate attention weights Through attention weight a i Adjust the normalized hidden layer output Hs to obtain the new hidden layer output H′; (2.5.5) Introduce dropout into the extreme learning machine and compute the updated hidden layer output H′; (2.5.6) Substitute H″ into the output layer calculation formula of the extreme learning machine to obtain the output layer output O.

9. The method for master data archiving processing based on Extreme Learning Machine in a domestically developed information technology environment according to claim 8, characterized in that, The calculation of the hidden layer output H in step (2.5.1) is specifically as follows: Calculate the hidden layer output H using the following formula: H = σ(XW + b); Where σ(·) is the activation function, X is the input feature matrix, and W and b are the weight matrix and bias vector of the hidden layer, respectively; In step (2.5.2), batch normalization of the hidden layer output H is performed, specifically as follows: Batch normalize the hidden layer output H according to the following formula: Where Hs is the normalized hidden layer output, μ H It is the mean of H. Let H be the variance of H, and ∈ be a constant. In step (2.5.3), the attention score s is calculated. i Specifically: Calculate the attention score s using the following formula. i : Among them, w a It is the attention weight, b a It's a bias, h i Let H be the i-th column, and tanh be the activation function; In step (2.5.4), attention weights are calculated. Specifically: Calculate the attention weight using the following formula. The calculation of the updated hidden layer output H′ in step (2.5.5) is specifically as follows: The updated hidden layer output H′ is calculated using the following formula: H″=D⊙H′; Where ⊙ represents the Hadamard product and D is the corresponding dropout mask; The calculation of the output layer output O in step (2.5.6) is specifically as follows: Calculate the output layer output O using the following formula: O = H″β; Where β is the output layer weight, and O is the classification result of the extreme learning machine.

10. An apparatus for implementing master data archiving processing based on Extreme Learning Machine in a domestic information technology innovation environment, characterized in that, The device includes: A processor is configured to execute computer-executable instructions; The memory stores one or more computer-executable instructions, which, when executed by the processor, implement the steps of the method for master data archiving processing based on an Extreme Learning Machine under the information technology innovation environment as described in any one of claims 1 to 9.

11. A processor for implementing master data archiving processing based on Extreme Learning Machine in a domestic IT innovation environment, characterized in that, The processor is configured to execute computer-executable instructions, which, when executed by the processor, implement the steps of the method for master data archiving processing based on Extreme Learning Machine under the information technology innovation environment as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, It stores a computer program that can be executed by a processor to implement the steps of the method for master data archiving processing based on Extreme Learning Machine in the information technology innovation environment as described in any one of claims 1 to 9.

Images