A big data management optimization method and system for data storage
Data identification is performed through dual-path compaction bidirectional long-term neural network and metaheuristic optimization algorithm, combined with symmetric key packet cryptographic encryption algorithm, the problem of resource waste, identification accuracy and security in big data management is solved, and efficient and secure Data storage management.
Patent Information
- Application Number
- CN202411174450.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-26
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-08-26
AI Technical Summary
The existing big data management optimization methods have problems in data storage such as wasted computing resources, reduced identification and classification accuracy, high model complexity and insufficient encryption security.
The two-path compaction bidirectional long and short-term neural network is used for data recognition, combined with the metaheuristic optimization algorithm to optimize model parameters, and the symmetric key packet cipher encryption algorithm is used to combine exclusive or encryption, modulo addition encryption and left-bit-based loop encryption for data encryption .
It saves computing resources in the data storage process, improves the depth and security of identification and classification, and optimizes the overall efficiency and security of data storage management.
Smart Images

Figure CN118673537B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of big data storage management, and in particular to a big data management optimization method and system for data storage. Background Art
[0002] A big data management optimization method for data storage refers to a method that aims to improve the efficiency, reliability and scalability of data management by adopting a series of technical means and strategies to address the challenges of large-scale data storage, processing and query. Its main purpose is to optimize the performance, resource utilization and cost-effectiveness of big data processing, which can help organizations manage large-scale data more effectively, improve the efficiency of data processing and analysis, and thus provide better decision support and business value for enterprises and organizations.
[0003] However, among the existing big data management optimization methods, there is a technical problem that the existing big data storage management optimization methods often adopt the traditional unified encryption scheme for security protection, but this method is easy to generate unnecessary public information encryption, which leads to the waste of computing resources and cloud space resources, and thus leads to the technical problem that the accuracy of the big data management method for data storage needs to be improved; among the existing big data management optimization methods, there is a technical problem that the existing big data input recognition method for data storage often performs simple binary classification, and the insufficient diversity of the classification dimensions leads to the reduction of the accuracy and reliability of the recognition classification, which leads to the technical problem that the artificial usability and management reference of the artificial intelligence method for data recognition are reduced; among the existing big data management optimization methods, there is a technical problem that the traditional method combining the compact convolutional network and the bidirectional long short-term memory model has a high model complexity, and there are a variety of data with complex dimensions, types and structures in data storage management, because the simple model construction cannot meet the needs of big data management, and the model performance needs to be improved; among the existing big data management optimization methods, there is a technical problem that the data storage for privacy data protection is vulnerable to spectrum analysis attacks when using XOR encryption alone, and the security is low, and the modular encryption method used alone is also vulnerable to frequency analysis attacks, and the use of a simple encryption algorithm cannot meet the needs of data storage. Summary of the invention
[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a big data management optimization method and system for data storage. In view of the fact that the existing big data management optimization methods often adopt the traditional unified encryption scheme for security protection, but this method is prone to unnecessary public information encryption, which leads to the waste of computing resources and cloud space resources, and thus leads to the technical problem that the accuracy of the big data management method for data storage needs to be improved, this scheme creatively adopts a dual-path tightened bidirectional long-term and short-term neural network to identify input data, distinguishes which data has a higher encryption priority, thereby saving computing resources in the data storage process and improving the efficiency and accuracy of data storage management; In view of the fact that the existing big data input identification method for data storage often performs simple binary classification, the classification dimension is not diverse enough, resulting in reduced accuracy and reliability of identification and classification, and thus leading to the technical problem that the artificial usability and management reference of artificial intelligence methods for data identification are reduced, this scheme creatively adopts a neural network method to classify data into five encryption priorities, improves the depth and dimension of identification and classification, and improves the interpretability of identification results and the overall usability of the method; In view of the fact that the existing big data management optimization methods often perform simple binary classification, the classification dimension is not diverse enough, resulting in reduced accuracy and reliability of identification and classification, and thus leads to the technical problem that the artificial usability and management reference of artificial intelligence methods for data identification are reduced, this scheme creatively adopts a neural network method to classify data into five encryption priorities, improves the depth and dimension of identification and classification, and improves the interpretability of identification results and the overall usability of the method; In the data storage management, there is a traditional method that combines a compact convolutional network and a bidirectional long short-term memory model, but the model complexity is high. In addition, there are data of various complex dimensions, types and structures in data storage management. Because simple model construction cannot meet the needs of big data management, and the model performance needs to be improved, this solution creatively adopts a dual-path compact bidirectional long short-term neural network combined with a meta-heuristic optimization algorithm. By constructing a dual-path compact bidirectional long short-term neural network, the specific objects of data recognition are differentiated from the model training stage, and the meta-heuristic optimization algorithm is used to optimize the model parameters, which improves the overall performance of the model and the overall optimization of data storage management. Effect: In the existing big data management optimization methods, there is a data storage for privacy data protection. When XOR encryption is used alone, it is vulnerable to spectrum analysis attacks and has low security. Modulo encryption alone is also vulnerable to frequency analysis attacks. The use of simple encryption algorithms cannot meet the needs of data storage. This solution creatively uses a symmetric key block cipher encryption algorithm, combines XOR encryption, modulo addition encryption and left bitwise cyclic encryption methods to encrypt data storage, thereby improving the security of data encryption, and solves the resource consumption caused by the high complexity of the encryption algorithm through pre-data classification, thereby optimizing the overall process of big data storage management.
[0005] The technical solution adopted by the present invention is as follows: The present invention provides a big data management optimization method for data storage, the method comprising the following steps:
[0006] Step S1: Data system construction;
[0007] Step S2: User information verification;
[0008] Step S3: input data identification;
[0009] Step S4: data encrypted storage;
[0010] Step S5: Big data management optimization.
[0011] Further, in step S1, the data system is constructed to construct a basic data storage system for big data management optimization, specifically, by constructing a big data management system for cloud environment data storage to obtain a basic data management subsystem;
[0012] The data management basic subsystem is used for data storage and big data management optimization, and specifically includes a user registration and authentication unit, a storage data identification unit and a data encryption unit.
[0013] Furthermore, in step S2, the user information verification is used to verify the basic information of the user logging into the system and activate the data identification and data encryption storage functions. Specifically, the user information is verified by constructing a user registration and authentication unit to obtain user information verification feedback information and data to be stored.
[0014] Further, in step S3, the input data recognition is used to identify and determine whether the user data contains sensitive data that needs to be encrypted through an artificial intelligence method. Specifically, based on the data to be stored, a dual-path compact bidirectional long-short term neural network combined with a meta-heuristic optimization algorithm is used to perform input data recognition to obtain data recognition data information, and the data recognition data information specifically includes sensitive data that needs to be encrypted and normal data that does not need to be encrypted;
[0015] The dual-path compact bidirectional long-short term neural network combined with a meta-heuristic optimization algorithm specifically includes a compact convolution sub-block, a bidirectional long-short term sub-block, a dense sub-block, an attention sub-block, a classifier sub-block and a meta-heuristic hyperparameter optimization operator;
[0016] The compact convolution sub-block is used to construct a compact convolutional neural network as an encoder and extract features;
[0017] The bidirectional long short-term sub-block is used to construct a bidirectional long short-term memory network as a decoder and optimize the model's processing capability for time series data;
[0018] The dense sub-block is used to integrate the output of the dual-path network sub-block and use it for the final classification prediction;
[0019] The attention sub-block is used to introduce an attention mechanism to optimize the feature extraction capability of the model;
[0020] The classifier sub-block is used to classify data types and identify input data;
[0021] The meta-heuristic hyperparameter optimization algorithm is used to optimize the model hyperparameters of the compact convolution sub-block;
[0022] The step of using a dual-path compact bidirectional long-short term neural network combined with a meta-heuristic optimization algorithm to identify input data and obtain data identification data information includes:
[0023] Step S31: training data collection and preprocessing, specifically, collecting statistical training sample data for input data recognition from a public data set, and performing data preprocessing operations on the statistical training data to obtain optimized training sample data;
[0024] The statistical training sample data specifically includes information sensitivity classification data;
[0025] The information sensitivity classification data specifically includes level zero sensitive data, level one sensitive data, level two sensitive data, level three sensitive data and level four sensitive data;
[0026] Step S32: constructing a dual-path compact bidirectional long-term and short-term subnetwork, including constructing a compact convolution sub-block and constructing a bidirectional long-term and short-term sub-block, specifically including the following steps:
[0027] Step S321: constructing a compressed convolution sub-block, specifically constructing a compressed convolution sub-block including a flame module, a nonlinear activation function, a convolution module and a maximum pooling module;
[0028] The specific construction steps of the compact convolution sub-block include:
[0029] Step S3211: constructing a first convolution module for processing and extracting data features of input data samples;
[0030] Step S3212: constructing a flame module, specifically constructing 8 flame modules in sequence for performing expansion and contraction operations, wherein the flame module specifically includes an input layer, a 1×1 convolution contraction layer, a 1×1 convolution expansion layer, a 3×3 convolution expansion layer, and a connection output layer;
[0031] Step S3213: a second convolution module, used to process the output features of the flame module and extract the compressed convolution output feature data;
[0032] Step S3214: construct a nonlinear activation function to introduce nonlinearity;
[0033] Step S3215: construct a maximum pooling module for performing feature upsampling through a pooling operation;
[0034] Step S3216: constructing a compressed convolution sub-block, specifically, constructing the compressed convolution sub-block according to the operations described in steps S3211 to S3215;
[0035] Step S322: construct a bidirectional long-term and short-term sub-block, specifically construct a bidirectional long-term and short-term neural network, as the bidirectional long-term and short-term sub-block, the calculation formula is:
[0036] ;
[0037] In the formula, F t is the output of the forget gate at time t, t is the time index, SIG(·) is the S-type activation function, is the hidden state weight of the forget gate, H t-1 is the hidden state at the previous time t-1, is the input weight of the forget gate, p t is the current input at time t, b F is the forget gate bias term, At is the input gate output, is the input gate hidden state weight, is the input gate input weight, b A is the input gate bias term, C t is the cell state data at time t, C t-1 is the cell state data at the previous time t-1, tanh(·) is the hyperbolic tangent function, is the cell state hidden state weight, is the cell state input weight, b C is the cell state bias term, Z t is the output gate output, is the output gate hidden state weight, is the output gate input weight, b O is the output gate bias term, H t is the hidden state at time t, and C is the updated cell state;
[0038] Step S323: constructing a dual-path compact bidirectional long-short term subnetwork, specifically, constructing a first path compact bidirectional long-short term subnetwork through the constructing compact convolution sub-block and the constructing bidirectional long-short term sub-block, and constructing a second path compact bidirectional long-short term subnetwork through the constructing compact convolution sub-block and the constructing bidirectional long-short term sub-block;
[0039] The first path compacts the bidirectional long-short term subnet for processing unstructured data;
[0040] The second path compacts the bidirectional long-short term subnet for processing structured data;
[0041] Step S33: constructing a dense sub-block, specifically, connecting the outputs of the first path-compacted bidirectional long-short term sub-network and the second path-compacted bidirectional long-short term sub-network to the dense sub-block, and constructing the dense sub-block by performing feature integration to obtain dual-path integrated feature data;
[0042] Step S34: constructing an attention sub-block, specifically introducing an attention function to mix attention features and obtain attention mixed feature data;
[0043] Step S35: constructing a classifier sub-block, specifically using a softmax classifier, performing input data sensitivity classification according to the attention mixed feature data, and obtaining input data sensitive identification information;
[0044] Step S36: Input data recognition model training, specifically, through the training data collection and preprocessing, the construction of the dual-path compact bidirectional long-term and short-term subnet, the construction of the dense sub-block, the construction of the attention sub-block and the construction of the classifier sub-block, the model training is performed to obtain the data recognition initial model Model DC , and obtain a model hyperparameter group through model training, wherein the model hyperparameter group specifically includes a learning rate, a batch size, a number of iterations, a loss function, and a deactivation rate;
[0045] Step S37: constructing a meta-heuristic hyperparameter optimization algorithm, specifically updating the hyperparameters through a rat colony optimization algorithm, including the following steps:
[0046] Step S371: Initialize the hyperparameter combination, the calculation formula is:
[0047] ;
[0048] Where X is the hyperparameter initialization combination, x1 is the first hyperparameter group, x2 is the second hyperparameter group, and x n is the nth hyperparameter group, n is the total number of hyperparameter combinations, Le1 is the learning rate parameter of the first hyperparameter group, Ba1 is the batch size parameter of the first hyperparameter group, Ep1 is the number of iterations parameter of the first hyperparameter group, Or1 is the loss function parameter of the first hyperparameter group, and Dr1 is the inactivation rate parameter of the first hyperparameter group;
[0049] Step S372: construct a fitness function, the calculation formula is:
[0050] Fit = Max(Accuracy);
[0051] Where Fit is the fitness function, Max(·) is the maximum value function, and Accuracy is the initial model Model for data recognition. DC Model accuracy;
[0052] Step S373: construct a hunting calculation function for calculating the optimal parameter position, and the calculation formula is:
[0053] ;
[0054] In the formula, is the optimal parameter position, x is the hyperparameter group index, is the individual linear scaling parameter, is the current individual optimal position vector, is the linear scaling factor for the entire population, is the global optimal position vector;
[0055] Step S374: construct a hunting calculation function for calculating the parameter optimization process, and the calculation formula is:
[0056] ;
[0057] In the formula, is the optimal solution of the optimization individual, |·| is the vector modulus operator, is the global optimal position vector, is the optimal parameter position;
[0058] Step S375: constructing an optimization termination condition, specifically setting the optimization termination condition to reach a maximum number of iterations or reach an optimal fitness value, and by constructing the optimization termination condition, obtaining an optimal hyperparameter combination when the optimization is terminated;
[0059] Step S376: Model optimization, specifically, using the meta-heuristic hyper-parameter optimization algorithm and combining the optimal hyper-parameter combination to perform the data recognition initial model Model DC Hyperparameter optimization to obtain the optimized data recognition model Model DC ;
[0060] Step S38: Input data recognition, specifically using the optimized data recognition model Model DC , input data identification is performed on the data to be stored to obtain data identification data information, the data identification data information specifically includes identification summary, sensitive data to be encrypted and normal data that does not need to be encrypted, the identification summary is used to describe the identification classification of the data, the sensitive data to be encrypted specifically includes first-level sensitive data, second-level sensitive data, third-level sensitive data and fourth-level sensitive data, the normal data that does not need to be encrypted specifically refers to zero-level sensitive data.
[0061] Further, in step S4, the data encryption storage is used to encrypt the sensitive data that needs to be encrypted, and store the encrypted data and the common data that does not need to be encrypted in the cloud storage of the data management basic subsystem, specifically using a symmetric key block cipher encryption algorithm to encrypt the data and obtain storage feedback information;
[0062] The storage feedback information specifically includes encrypted storage feedback and unencrypted storage feedback;
[0063] The symmetric key block cipher encryption algorithm specifically includes XOR encryption, modulo addition encryption and left bitwise circular encryption;
[0064] The step of using a symmetric key block cipher encryption algorithm to encrypt and store data to obtain storage feedback information includes:
[0065] Step S41: Construct an XOR encryption block, the calculation formula is:
[0066] ;
[0067] In the formula, Code X is the XOR encrypted ciphertext, PT is the source data plaintext, Key X is the XOR encryption key;
[0068] Step S42: construct a modulo addition encryption block, the calculation formula is:
[0069] ;
[0070] In the formula, Code M is the modulo addition encryption ciphertext, PT is the source data plaintext, Key M is the modulo addition encryption key, Y is the modulo parameter, and mod is the modulo operator;
[0071] Step S43: construct a left bitwise circular encryption block, the calculation formula is:
[0072] ;
[0073] In the formula, Code R is the left bitwise circular encrypted ciphertext, PT is the source data plaintext, and N is the number of shift bits;
[0074] Step S44: encrypting the ciphertext, specifically, by constructing the XOR encryption block, constructing the modulo addition encryption block, and constructing the left bitwise circular encryption block, using a 64-bit text block and a 128-bit key, and performing 32 rounds of ciphertext encryption through the Feistel network to obtain encrypted ciphertext data;
[0075] Step S45: constructing a decryption block, specifically, by using the key according to the operations described in steps S41 to S43, restoring the ciphertext to the source data plaintext, decrypting the ciphertext, and constructing the decryption block, wherein the decryption block is specifically used to read the stored encrypted data;
[0076] Step S46: data storage, specifically, storing the encrypted ciphertext data to cloud storage to obtain encrypted storage feedback, and storing the normal data that does not need to be encrypted to cloud storage to obtain unencrypted storage feedback, completing data storage and obtaining storage feedback information.
[0077] Further, in step S5, the big data management optimization is used to perform big data management optimization of data storage through data identification and encrypted storage, specifically by obtaining the user information verification feedback information, combining the identification summary in the data identification data information, and performing big data management optimization based on the storage feedback information to obtain data storage management feedback information.
[0078] A big data management optimization system for data storage provided by the present invention comprises a data system construction module, a user information verification module, an input data identification module, a data encryption storage module and a big data management optimization module;
[0079] The data system construction module is used for data system construction. Through the data system construction, a data management basic subsystem is obtained, and the data management basic subsystem is used for user information verification, input data identification and data encryption storage;
[0080] The user information verification module is used for user information verification, obtains user information verification feedback information and data to be stored through user information verification, sends the data to be stored to the input data identification module, and sends the user information verification feedback information to the big data management optimization module;
[0081] The input data identification module is used for input data identification, obtains data identification data information through input data identification, and sends the data identification data information to the data encryption storage module and the big data management optimization module;
[0082] The data encryption storage module is used for data encryption storage, obtains storage feedback information through data encryption storage, and sends the storage feedback information to the big data management optimization module;
[0083] The big data management optimization module is used for big data management optimization, and data storage management feedback information is obtained through big data management optimization.
[0084] The beneficial effects achieved by the present invention using the above scheme are as follows:
[0085] (1) In view of the fact that the existing big data management optimization methods often adopt the traditional unified encryption scheme for security protection, this method is prone to unnecessary encryption of public information, which leads to the waste of computing resources and cloud space resources, and thus leads to the technical problem that the accuracy of big data management methods for data storage needs to be improved. This scheme creatively adopts a dual-path tightened bidirectional long-short-term neural network to identify input data, distinguish which data has a higher encryption priority, thereby saving computing resources in the data storage process and improving the efficiency and accuracy of data storage management;
[0086] (2) In view of the technical problem that the existing big data management optimization methods often use simple binary classification for data storage input recognition methods, the lack of diversity in the classification dimensions leads to reduced accuracy and reliability of recognition and classification, which in turn leads to reduced human usability and management reference of artificial intelligence methods for data recognition. This solution creatively uses a neural network method to classify data into five encryption priorities, which improves the depth and dimension of recognition and classification, and improves the interpretability of recognition results and the overall usability of the method.
[0087] (3) In view of the fact that among the existing big data management optimization methods, the traditional method combining the compact convolutional network and the bidirectional long short-term memory model has a high model complexity, and there are data of various complex dimensions, types and structures in data storage management, and the simple model construction cannot meet the needs of big data management, and the model performance needs to be improved, this solution creatively adopts a dual-path compact bidirectional long short-term neural network combined with a meta-heuristic optimization algorithm. By constructing a dual-path compact bidirectional long short-term neural network, the specific objects of data recognition are differentiated from the model training stage, and the meta-heuristic optimization algorithm is used to optimize the model parameters, thereby improving the overall performance of the model and the overall optimization effect of data storage management;
[0088] (4) In view of the technical problem that in the existing big data management optimization methods, there is data storage for privacy data protection. When using XOR encryption alone, it is vulnerable to spectrum analysis attacks and has low security. When using modulo encryption alone, it is also vulnerable to frequency analysis attacks. The use of simple encryption algorithms cannot meet the needs of data storage. This solution creatively adopts a symmetric key block cipher encryption algorithm, combines XOR encryption, modulo addition encryption and left bitwise cyclic encryption methods to encrypt data storage, thereby improving the security of data encryption. It also solves the resource consumption caused by the high complexity of the encryption algorithm through pre-data classification, and optimizes the overall process of big data storage management. BRIEF DESCRIPTION OF THE DRAWINGS
[0089] Figure 1A schematic diagram of a flow chart of a big data management optimization method for data storage provided by the present invention;
[0090] Figure 2 A schematic diagram of a big data management optimization system for data storage provided by the present invention;
[0091] Figure 3 This is a flow chart of user information verification in step S2;
[0092] Figure 4 A schematic diagram of the process of inputting data identification in step S3;
[0093] Figure 5 This is a schematic diagram of the process of data encryption and storage in step S4;
[0094] Figure 6 A visualization diagram of the identification summary of the data information identified in step S3.
[0095] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION
[0096] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0097] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.
[0098] Example 1, see Figure 1 The present invention provides a big data management optimization method for data storage, the method comprising the following steps:
[0099] Step S1: Data system construction;
[0100] Step S2: User information verification;
[0101] Step S3: input data identification;
[0102] Step S4: data encrypted storage;
[0103] Step S5: Big data management optimization.
[0104] Example 2, see Figure 1 and Figure 2 In step S1, the data system is constructed to construct a basic data storage system for big data management optimization, specifically, by constructing a big data management system for cloud environment data storage to obtain a basic data management subsystem;
[0105] The data management basic subsystem is used for data storage and big data management optimization, and specifically includes a user registration and authentication unit, a storage data identification unit, and a data encryption unit;
[0106] The user registration and authentication unit is used for user information verification;
[0107] The stored data identification unit is used for input data identification;
[0108] The data encryption unit is used for data encryption storage.
[0109] Example 3, see Figure 1 , Figure 2 and Figure 3 ,This embodiment is based on the above embodiment. In step S2, in step S2, the user information verification is used to verify the basic information of the user who logs into the system and activate the data identification and data encryption storage functions. Specifically, by constructing a user registration and authentication unit, the user information verification is performed to obtain user information verification feedback information and data to be stored;
[0110] The user registration and authentication unit specifically includes a user registration unit and a user authentication unit;
[0111] The user registration unit is used to create user account information and generate a key;
[0112] The user authentication unit is used to verify user account information and perform system access;
[0113] The step of verifying user information by constructing a user registration and authentication unit includes:
[0114] Step S21: system authorization, specifically, the user accesses the data management basic subsystem, performs system authorization through the data management basic subsystem, and starts the user information registration program to create a user personal account;
[0115] Step S22: user registration, specifically, by constructing the user registration unit, sending a personal key to the user's personal account, and the user logs in to the user's personal account by receiving the personal key;
[0116] Step S23: user authentication, specifically, by constructing the user authentication unit to perform user identity information verification, the user identity information verification specifically includes user account identifier verification, user password verification, random number verification and personal key verification;
[0117] Step S24: user login, specifically, through the user registration and the user authentication, the user logs in to the user account, the user uploads the data to be stored by logging in, and the data management basic subsystem obtains user information verification feedback information by logging in;
[0118] The data to be stored is used for input data identification;
[0119] The user information verification feedback information is used for big data management optimization.
[0120] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the input data recognition is used to identify and determine whether the user data contains sensitive data that needs to be encrypted through an artificial intelligence method. Specifically, based on the data to be stored, a dual-path compact bidirectional long-term and short-term neural network combined with a meta-heuristic optimization algorithm is used to perform input data recognition to obtain data recognition data information. The data recognition data information specifically includes sensitive data that needs to be encrypted and normal data that does not need to be encrypted;
[0121] The dual-path compact bidirectional long-short term neural network combined with a meta-heuristic optimization algorithm specifically includes a compact convolution sub-block, a bidirectional long-short term sub-block, a dense sub-block, an attention sub-block, a classifier sub-block and a meta-heuristic hyperparameter optimization operator;
[0122] The compact convolution sub-block is used to construct a compact convolutional neural network as an encoder and extract features;
[0123] The bidirectional long short-term sub-block is used to construct a bidirectional long short-term memory network as a decoder and optimize the model's processing capability for time series data;
[0124] The dense sub-block is used to integrate the output of the dual-path network sub-block and use it for the final classification prediction;
[0125] The attention sub-block is used to introduce an attention mechanism to optimize the feature extraction capability of the model;
[0126] The classifier sub-block is used to classify data types and identify input data;
[0127] The meta-heuristic hyperparameter optimization algorithm is used to optimize the model hyperparameters of the compact convolution sub-block;
[0128] By performing the above operations, in view of the fact that in the existing big data management optimization methods, the existing big data storage management optimization methods often adopt traditional unified encryption schemes for security protection, but this method is prone to unnecessary public information encryption, which leads to waste of computing resources and cloud space resources, and further leads to the technical problem that the accuracy of big data management methods for data storage needs to be improved. This scheme creatively adopts a dual-path tightened bidirectional long-short-term neural network to identify input data, distinguishes which data has a higher encryption priority, thereby saving computing resources in the data storage process and improving the efficiency and accuracy of data storage management.
[0129] Example 5, see Figure 1 , Figure 2 , Figure 4 and Figure 6 In step S3, the step of using a dual-path compact bidirectional long-short term neural network combined with a meta-heuristic optimization algorithm to identify input data and obtain data identification data information includes:
[0130] Step S31: training data collection and preprocessing, specifically, collecting statistical training sample data for input data recognition from a public data set, and performing data preprocessing operations on the statistical training data to obtain optimized training sample data;
[0131] The statistical training sample data specifically includes information sensitivity classification data;
[0132] The information sensitivity classification data specifically includes level zero sensitive data, level one sensitive data, level two sensitive data, level three sensitive data and level four sensitive data;
[0133] The zero-level sensitive data refers specifically to non-private data that can be obtained by the public through public channels;
[0134] The first-level sensitive data specifically refers to the user's basic personal information data;
[0135] The second level of sensitive data refers specifically to user detailed personal privacy information data;
[0136] The third-level sensitive data refers specifically to basic information data of enterprises or institutions;
[0137] The fourth level of sensitive data refers specifically to the private information data of enterprises or institutions;
[0138] The data preprocessing operation specifically includes data normalization, missing value processing and data standardization operations;
[0139] Step S32: constructing a dual-path compact bidirectional long-term and short-term subnetwork, including constructing a compact convolution sub-block and constructing a bidirectional long-term and short-term sub-block, specifically including the following steps:
[0140] Step S321: constructing a compressed convolution sub-block, specifically constructing a compressed convolution sub-block including a flame module, a nonlinear activation function, a convolution module and a maximum pooling module;
[0141] The specific construction steps of the compact convolution sub-block include:
[0142] Step S3211: constructing a first convolution module for processing and extracting data features of input data samples;
[0143] Step S3212: constructing a flame module, specifically constructing 8 flame modules in sequence for performing expansion and contraction operations, wherein the flame module specifically includes an input layer, a 1×1 convolution contraction layer, a 1×1 convolution expansion layer, a 3×3 convolution expansion layer, and a connection output layer;
[0144] Step S3213: a second convolution module, used to process the output features of the flame module and extract the compressed convolution output feature data;
[0145] Step S3214: construct a nonlinear activation function to introduce nonlinearity;
[0146] Step S3215: construct a maximum pooling module for performing feature upsampling through a pooling operation;
[0147] Step S3216: constructing a compressed convolution sub-block, specifically, constructing the compressed convolution sub-block according to the operations described in steps S3211 to S3215;
[0148] Step S322: construct a bidirectional long-term and short-term sub-block, specifically construct a bidirectional long-term and short-term neural network, as the bidirectional long-term and short-term sub-block, the calculation formula is:
[0149] ;
[0150] In the formula, F t is the output of the forget gate at time t, t is the time index, SIG(·) is the S-type activation function, is the hidden state weight of the forget gate, H t-1 is the hidden state at the previous time t-1, is the input weight of the forget gate, p t is the current input at time t, b F is the forget gate bias term, At is the input gate output, is the input gate hidden state weight, is the input gate input weight, b A is the input gate bias term, C t is the cell state data at time t, C t-1 is the cell state data at the previous time t-1, tanh(·) is the hyperbolic tangent function, is the cell state hidden state weight, is the cell state input weight, b C is the cell state bias term, Z t is the output gate output, is the output gate hidden state weight, is the output gate input weight, b O is the output gate bias term, H t is the hidden state at time t, and C is the updated cell state;
[0151] Step S323: constructing a dual-path compact bidirectional long-short term subnetwork, specifically, constructing a first path compact bidirectional long-short term subnetwork through the constructing compact convolution sub-block and the constructing bidirectional long-short term sub-block, and constructing a second path compact bidirectional long-short term subnetwork through the constructing compact convolution sub-block and the constructing bidirectional long-short term sub-block;
[0152] The first path compacts the bidirectional long-short term subnet for processing unstructured data;
[0153] The second path compacts the bidirectional long-short term subnet for processing structured data;
[0154] Step S33: constructing a dense sub-block, specifically, connecting the outputs of the first path-compacted bidirectional long-short term sub-network and the second path-compacted bidirectional long-short term sub-network to the dense sub-block, and constructing the dense sub-block by performing feature integration to obtain dual-path integrated feature data;
[0155] Step S34: constructing an attention sub-block, specifically introducing an attention function to mix attention features and obtain attention mixed feature data;
[0156] Step S35: constructing a classifier sub-block, specifically using a softmax classifier, performing input data sensitivity classification according to the attention mixed feature data, and obtaining input data sensitive identification information;
[0157] Step S36: Input data recognition model training, specifically, through the training data collection and preprocessing, the construction of the dual-path compact bidirectional long-term and short-term subnet, the construction of the dense sub-block, the construction of the attention sub-block and the construction of the classifier sub-block, the model training is performed to obtain the data recognition initial model Model DC, and obtain a model hyperparameter group through model training, wherein the model hyperparameter group specifically includes a learning rate, a batch size, a number of iterations, a loss function, and a deactivation rate;
[0158] By performing the above operations, in view of the technical problem that in the existing big data management optimization methods, the existing big data input recognition methods for data storage often perform simple binary classification, and the insufficient diversity of the classification dimensions leads to reduced accuracy and reliability of recognition and classification, which in turn leads to reduced manual usability and management reference of artificial intelligence methods for data recognition, this solution creatively adopts a neural network method to classify data into five encryption priorities, thereby improving the depth and dimension of recognition and classification, and improving the interpretability of recognition results and the overall usability of the method;
[0159] Step S37: constructing a meta-heuristic hyperparameter optimization algorithm, specifically updating the hyperparameters through a rat colony optimization algorithm, including the following steps:
[0160] Step S371: Initialize the hyperparameter combination, the calculation formula is:
[0161] ;
[0162] Where X is the hyperparameter initialization combination, x1 is the first hyperparameter group, x2 is the second hyperparameter group, and x n is the nth hyperparameter group, n is the total number of hyperparameter combinations, Le1 is the learning rate parameter of the first hyperparameter group, Ba1 is the batch size parameter of the first hyperparameter group, Ep1 is the number of iterations parameter of the first hyperparameter group, Or1 is the loss function parameter of the first hyperparameter group, and Dr1 is the inactivation rate parameter of the first hyperparameter group;
[0163] Step S372: construct a fitness function, the calculation formula is:
[0164] Fit = Max(Accuracy);
[0165] Where Fit is the fitness function, Max(·) is the maximum value function, and Accuracy is the initial model Model for data recognition. DC Model accuracy;
[0166] Step S373: construct a hunting calculation function for calculating the optimal parameter position, and the calculation formula is:
[0167] ;
[0168] In the formula, is the optimal parameter position, x is the hyperparameter group index, is the individual linear scaling parameter, is the current individual optimal position vector, is the linear scaling factor for the entire population, is the global optimal position vector;
[0169] Step S374: construct a hunting calculation function for calculating the parameter optimization process, and the calculation formula is:
[0170] ;
[0171] In the formula, is the optimal solution of the optimization individual, |·| is the vector modulus operator, is the global optimal position vector, is the optimal parameter position;
[0172] Step S375: constructing an optimization termination condition, specifically setting the optimization termination condition to reach a maximum number of iterations or reach an optimal fitness value, and by constructing the optimization termination condition, obtaining an optimal hyperparameter combination when the optimization is terminated;
[0173] Step S376: Model optimization, specifically, using the meta-heuristic hyper-parameter optimization algorithm and combining the optimal hyper-parameter combination to perform the data recognition initial model Model DC Hyperparameter optimization to obtain the optimized data recognition model Model DC ;
[0174] Step S38: Input data recognition, specifically using the optimized data recognition model Model DC , performing input data identification on the data to be stored to obtain data identification data information, wherein the data identification data information specifically includes identification summary, sensitive data to be encrypted and general data not to be encrypted, wherein the identification summary is used to describe the identification classification of the data, wherein the sensitive data to be encrypted specifically includes first-level sensitive data, second-level sensitive data, third-level sensitive data and fourth-level sensitive data, and the general data not to be encrypted specifically refers to zero-level sensitive data;
[0175] See also Figure 6 , where the horizontal coordinate, represents the level of sensitive data in the identification summary, where the value 0 corresponds to level zero sensitive data, the value 1 corresponds to level one sensitive data, the value 2 corresponds to level two sensitive data, the value 3 corresponds to level three sensitive data, and the value 4 corresponds to level four sensitive data;
[0176] The vertical coordinate indicates the number of specific identification data corresponding to each sensitive data level value. The number of specific identification data of level 0 sensitive data in the identification summary described this time is 12893, the number of specific identification data of level 1 sensitive data in the identification summary described this time is 8383, the number of specific identification data of level 2 sensitive data in the identification summary described this time is 10771, the number of specific identification data of level 3 sensitive data in the identification summary described this time is 5852, and the number of specific identification data of level 4 sensitive data in the identification summary described this time is 1327;
[0177] By performing the above operations, in view of the fact that in the existing big data management optimization methods, the traditional method combining the compact convolutional network and the bidirectional long short-term memory model has a high model complexity, and there are data of various complex dimensions, types and structures in data storage management, because simple model construction cannot meet the needs of big data management, and the model performance needs to be improved, this solution creatively adopts a dual-path compact bidirectional long short-term neural network combined with a meta-heuristic optimization algorithm. By constructing a dual-path compact bidirectional long short-term neural network, the specific objects of data recognition are differentiated from the model training stage, and the meta-heuristic optimization algorithm is used to optimize the model parameters, thereby improving the overall performance of the model and the overall optimization effect of data storage management.
[0178] Example 6, see Figure 1 , Figure 2 , Figure 3 and Figure 5 This embodiment is based on the above embodiment. In step S4, the data encryption storage is used to encrypt the sensitive data that needs to be encrypted, and store the encrypted data and the common data that does not need to be encrypted in the cloud storage of the data management basic subsystem. Specifically, a symmetric key block cipher encryption algorithm is used to encrypt the data and store the storage feedback information;
[0179] The storage feedback information specifically includes encrypted storage feedback and unencrypted storage feedback;
[0180] The symmetric key block cipher encryption algorithm specifically includes XOR encryption, modulo addition encryption and left bitwise circular encryption;
[0181] The step of using a symmetric key block cipher encryption algorithm to encrypt and store data to obtain storage feedback information includes:
[0182] Step S41: Construct an XOR encryption block, the calculation formula is:
[0183] ;
[0184] In the formula, Code Xis the XOR encrypted ciphertext, PT is the source data plaintext, Key X is the XOR encryption key;
[0185] Step S42: construct a modulo addition encryption block, the calculation formula is:
[0186] ;
[0187] In the formula, Code M is the modulo addition encryption ciphertext, PT is the source data plaintext, Key M is the modulo addition encryption key, Y is the modulo parameter, and mod is the modulo operator;
[0188] Step S43: construct a left bitwise circular encryption block, the calculation formula is:
[0189] ;
[0190] In the formula, Code R is the left bitwise circular encrypted ciphertext, PT is the source data plaintext, and N is the number of shift bits;
[0191] Step S44: encrypting the ciphertext, specifically, by constructing the XOR encryption block, constructing the modulo addition encryption block, and constructing the left bitwise circular encryption block, using a 64-bit text block and a 128-bit key, and performing 32 rounds of ciphertext encryption through the Feistel network to obtain encrypted ciphertext data;
[0192] Step S45: constructing a decryption block, specifically, by using the key according to the operations described in steps S41 to S43, restoring the ciphertext to the source data plaintext, decrypting the ciphertext, and constructing the decryption block, wherein the decryption block is specifically used to read the stored encrypted data;
[0193] Step S46: data storage, specifically, storing the encrypted ciphertext data to cloud storage to obtain encrypted storage feedback, and storing the normal data that does not need to be encrypted to cloud storage to obtain unencrypted storage feedback, completing data storage and obtaining storage feedback information.
[0194] By performing the above operations, in order to solve the technical problem that in the existing big data management optimization methods, there is data storage for privacy data protection, which is vulnerable to spectrum analysis attacks when using XOR encryption alone and has low security, and is also vulnerable to frequency analysis attacks when using modulo encryption alone, and the use of simple encryption algorithms cannot meet the needs of data storage, this solution creatively adopts a symmetric key block cipher encryption algorithm, combines XOR encryption, modulo addition encryption and left bitwise cyclic encryption methods to encrypt data storage, thereby improving the security of data encryption, and solves the resource consumption caused by the high complexity of the encryption algorithm through pre-data classification, thereby optimizing the overall process of big data storage management.
[0195] Embodiment 7, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the big data management optimization is used to optimize the big data management of data storage through data identification and encrypted storage. Specifically, by obtaining the user information verification feedback information, combining the identification summary in the data identification data information, and based on the storage feedback information, big data management optimization is performed to obtain data storage management feedback information.
[0196] Embodiment 8, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and a big data management optimization system for data storage provided by the present invention includes a data system construction module, a user information verification module, an input data identification module, a data encryption storage module and a big data management optimization module;
[0197] The data system construction module is used for data system construction. Through the data system construction, a data management basic subsystem is obtained, and the data management basic subsystem is used for user information verification, input data identification and data encryption storage;
[0198] The user information verification module is used for user information verification, obtains user information verification feedback information and data to be stored through user information verification, sends the data to be stored to the input data identification module, and sends the user information verification feedback information to the big data management optimization module;
[0199] The input data identification module is used for input data identification, obtains data identification data information through input data identification, and sends the data identification data information to the data encryption storage module and the big data management optimization module;
[0200] The data encryption storage module is used for data encryption storage, obtains storage feedback information through data encryption storage, and sends the storage feedback information to the big data management optimization module;
[0201] The big data management optimization module is used for big data management optimization, and data storage management feedback information is obtained through big data management optimization.
[0202] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0203] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.
[0204] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.
Claims
1. A big data management optimization method for data storage, Features: The method comprises the following steps: Step S1: Data system construction, which is used to construct a basic data storage system for big data management optimization, specifically, by constructing a big data management system for cloud environment data storage to obtain a basic data management subsystem; Step S2: User information verification, which is used to verify the basic information of the user who logs into the system and activate the data identification and data encryption storage functions. Specifically, the user information is verified by building a user registration and authentication unit to obtain user information verification feedback information and data to be stored; Step S3: input data identification, used to identify and determine whether the user data contains sensitive data that needs to be encrypted through an artificial intelligence method, specifically, based on the data to be stored, a dual-path compact bidirectional long-short term neural network combined with a meta-heuristic optimization algorithm is used to perform input data identification to obtain data identification data information, wherein the data identification data information specifically includes an identification summary, sensitive data that needs to be encrypted, and normal data that does not need to be encrypted. The identification summary is used to describe the identification classification of the data. The sensitive data that needs to be encrypted specifically includes first-level sensitive data, second-level sensitive data, third-level sensitive data, and fourth-level sensitive data. The normal data that does not need to be encrypted specifically refers to zero-level sensitive data; The dual-path compact bidirectional long-short term neural network combined with a meta-heuristic optimization algorithm specifically includes a compact convolution sub-block, a bidirectional long-short term sub-block, a dense sub-block, an attention sub-block, a classifier sub-block and a meta-heuristic hyperparameter optimization operator; The step of using a dual-path compact bidirectional long-short term neural network combined with a meta-heuristic optimization algorithm to identify input data and obtain data identification data information includes: Step S31: training data collection and preprocessing, specifically, collecting statistical training sample data for input data recognition from public data sets, and performing data preprocessing operations on the statistical training data to obtain optimized training sample data; The statistical training sample data specifically includes information sensitivity classification data; The information sensitivity classification data specifically includes level zero sensitive data, level one sensitive data, level two sensitive data, level three sensitive data and level four sensitive data; Step S32: constructing a dual-path compact bidirectional long-short term subnetwork, including constructing a compact convolution sub-block and constructing a bidirectional long-short term sub-block, to obtain a first-path compact bidirectional long-short term subnetwork and a second-path compact bidirectional long-short term subnetwork; The first path compacts the bidirectional long-short term subnet for processing unstructured data; The second path compacts the bidirectional long-short term subnet for processing structured data; Step S33: constructing a dense sub-block, specifically, connecting the outputs of the first path-compacted bidirectional long-short term sub-network and the second path-compacted bidirectional long-short term sub-network to the dense sub-block, and constructing the dense sub-block by performing feature integration to obtain dual-path integrated feature data; Step S34: constructing an attention sub-block, specifically introducing an attention function to mix attention features and obtain attention mixed feature data; Step S35: constructing a classifier sub-block, specifically using a softmax classifier, performing input data sensitivity classification according to the attention mixed feature data, and obtaining input data sensitive identification information; Step S36: Input data recognition model training, specifically, through the training data collection and preprocessing, the construction of the dual-path compact bidirectional long-term and short-term subnet, the construction of the dense sub-block, the construction of the attention sub-block and the construction of the classifier sub-block, the model training is performed to obtain the data recognition initial model Model DC , and obtain a model hyperparameter group through model training, wherein the model hyperparameter group specifically includes a learning rate, a batch size, a number of iterations, a loss function, and a deactivation rate; Step S37: Construct a meta-heuristic hyperparameter optimization algorithm, specifically update the hyperparameters through the mouse colony optimization algorithm to obtain the optimized data recognition model Model DC ; Step S38: Input data recognition, specifically using the optimized data recognition model Model DC , performing input data identification on the data to be stored to obtain data identification data information, wherein the data identification data information specifically includes identification summary, sensitive data to be encrypted and general data not to be encrypted, wherein the identification summary is used to describe the identification classification of the data, wherein the sensitive data to be encrypted specifically includes first-level sensitive data, second-level sensitive data, third-level sensitive data and fourth-level sensitive data, and the general data not to be encrypted specifically refers to zero-level sensitive data; Step S4: data encryption storage, which is used to encrypt sensitive data that needs to be encrypted, and store the encrypted data and common data that does not need to be encrypted in the cloud storage of the data management basic subsystem, specifically using a symmetric key block cipher encryption algorithm to encrypt the data and obtain storage feedback information; the storage feedback information specifically includes encrypted storage feedback and unencrypted storage feedback; The symmetric key block cipher encryption algorithm specifically includes XOR encryption, modulo addition encryption and left bitwise circular encryption; The step of using a symmetric key block cipher encryption algorithm to encrypt and store data to obtain storage feedback information includes: Step S41: Construct an XOR encryption block, the calculation formula is: In the formula, Code X is the XOR encrypted ciphertext, PT is the source data plaintext, Key X is the XOR encryption key; Step S42: construct a modulo addition encryption block, the calculation formula is: Code M =(PT+Key M )mod Y; In the formula, Code M is the modulo addition encryption ciphertext, PT is the source data plaintext, Key M is the modulo addition encryption key, Y is the modulo parameter, and mod is the modulo operator; Step S43: construct a left bitwise circular encryption block, the calculation formula is: Code R =(PT<<N)|(PT>>(8-N)); In the formula, Code R is the left bitwise circular encrypted ciphertext, PT is the source data plaintext, and N is the number of shift bits; Step S44: encrypting the ciphertext, specifically, by constructing the XOR encryption block, constructing the modulo addition encryption block, and constructing the left bitwise circular encryption block, using a 64-bit text block and a 128-bit key, and performing 32 rounds of ciphertext encryption through the Feistel network to obtain encrypted ciphertext data; Step S45: constructing a decryption block, specifically, by using the key according to the operations described in steps S41 to S43, restoring the ciphertext to the source data plaintext, decrypting the ciphertext, and constructing the decryption block, wherein the decryption block is specifically used to read the stored encrypted data; Step S46: data storage, specifically, storing the encrypted ciphertext data in the cloud storage to obtain encrypted storage feedback, and storing the normal data that does not need to be encrypted in the cloud storage to obtain unencrypted storage feedback, completing data storage and obtaining storage feedback information; Step S5: Big data management optimization, which is used to optimize the big data management of data storage through data identification and encrypted storage, and obtain data storage management feedback information.
2. A big data management optimization method for data storage according to claim 1, Features: In step S3, the dual-path compact bidirectional long-short term neural network combined with the meta-heuristic optimization algorithm specifically includes a compact convolution sub-block, a bidirectional long-short term sub-block, a dense sub-block, an attention sub-block, a classifier sub-block and a meta-heuristic hyperparameter optimization operator; The compact convolution sub-block is used to construct a compact convolutional neural network as an encoder and extract features; The bidirectional long short-term sub-block is used to construct a bidirectional long short-term memory network as a decoder and optimize the model's processing capability for time series data; The dense sub-block is used to integrate the output of the dual-path network sub-block and use it for the final classification prediction; The attention sub-block is used to introduce an attention mechanism to optimize the feature extraction capability of the model; The classifier sub-block is used to classify data types and identify input data; The meta-heuristic hyperparameter optimization algorithm is used to optimize the model hyperparameters of the compact convolution sub-block.
3. A big data management optimization method for data storage according to claim 2, Features: In step S32, the construction of the dual-path compact bidirectional long-short term subnetwork includes constructing a compact convolution sub-block and constructing a bidirectional long-short term sub-block, specifically including the following steps: Step S321: constructing a compressed convolution sub-block, specifically constructing a compressed convolution sub-block including a flame module, a nonlinear activation function, a convolution module and a maximum pooling module; The specific construction steps of the compact convolution sub-block include: Step S3211: constructing a first convolution module for processing and extracting data features of input data samples; Step S3212: constructing a flame module, specifically constructing 8 flame modules in sequence for performing expansion and contraction operations, wherein the flame module specifically includes an input layer, a 1×1 convolution contraction layer, a 1×1 convolution expansion layer, a 3×3 convolution expansion layer, and a connection output layer; Step S3213: a second convolution module, used to process the output features of the flame module and extract the compressed convolution output feature data; Step S3214: construct a nonlinear activation function to introduce nonlinearity; Step S3215: construct a maximum pooling module for performing feature upsampling through a pooling operation; Step S3216: constructing a compressed convolution sub-block, specifically, constructing the compressed convolution sub-block according to the operations of step S3211 to step S3215; Step S322: construct a bidirectional long-term and short-term sub-block, specifically construct a bidirectional long-term and short-term neural network, as the bidirectional long-term and short-term sub-block, the calculation formula is: In the formula, F t is the output of the forget gate at time t, t is the time index, SIG(·) is the S-type activation function, ω FH is the hidden state weight of the forget gate, H t-1 is the hidden state at the previous time t-1, ω Fp is the input weight of the forget gate, p t is the current input at time t, b F is the forget gate bias term, At is the input gate output, ω AH is the input gate hidden state weight, ω Ap is the input gate input weight, b A is the input gate bias term, C t is the cell state data at time t, C t-1 is the cell state data at the previous time t-1, tanh(·) is the hyperbolic tangent function, ω CH is the cell state hidden state weight, ω Cp is the cell state input weight, b C is the cell state bias term, Z t is the output gate output, ω ZH is the output gate hidden state weight, ω Zp is the output gate input weight, b O is the output gate bias term, H t is the hidden state at time t, and C is the updated cell state; Step S323: constructing a dual-path compact bidirectional long-short term subnetwork, specifically, constructing a first path compact bidirectional long-short term subnetwork through the constructing compact convolution sub-block and the constructing bidirectional long-short term sub-block, and constructing a second path compact bidirectional long-short term subnetwork through the constructing compact convolution sub-block and the constructing bidirectional long-short term sub-block.
4. A big data management optimization method for data storage according to claim 3, Features: In step S37, the construction of the meta-heuristic hyper-parameter optimization algorithm specifically refers to updating the hyper-parameters by using a mouse colony optimization algorithm. The following steps are involved: Step S371: Initialize the hyperparameter combination, the calculation formula is: Where X is the hyperparameter initialization combination, x 1 is the first hyperparameter group, x 2 is the second hyperparameter group, x n is the nth hyperparameter group, n is the total number of hyperparameter combinations, Le 1 is the learning rate parameter of the first hyperparameter group, Ba 1 is the first hyperparameter group batch size parameter, Ep 1 is the number of iterations of the first hyperparameter group, Or 1 is the loss function parameter of the first hyperparameter group, Dr 1 is the dropout rate parameter of the first hyperparameter group; Step S372: construct a fitness function, the calculation formula is: Fit = Max(Accuracy); Where Fit is the fitness function, Max(·) is the maximum value function, and Accuracy is the initial model Model for data recognition. DC Model accuracy; Step S373: construct a hunting calculation function for calculating the optimal parameter position, and the calculation formula is: In the formula, is the optimal parameter position, x is the hyperparameter group index, A′ is the individual linear scaling parameter, is the optimal position vector of the current individual, C′ is the linear scaling factor of the entire population, is the global optimal position vector; Step S374: construct a hunting calculation function for calculating the parameter optimization process, and the calculation formula is: In the formula, is the optimal solution of the optimization individual, |·| is the vector modulus operator, is the global optimal position vector, is the optimal parameter position; Step S375: constructing an optimization termination condition, specifically setting the optimization termination condition to reach a maximum number of iterations or reach an optimal fitness value, and by constructing the optimization termination condition, obtaining an optimal hyperparameter combination when the optimization is terminated; Step S376: Model optimization, specifically, using the meta-heuristic hyper-parameter optimization algorithm and combining the optimal hyper-parameter combination to perform the data recognition initial model Model DC Hyperparameter optimization to obtain the optimized data recognition model Model DC .
5. A big data management optimization method for data storage according to claim 4, Features: In step S5, the big data management optimization is used to perform big data management optimization of data storage through data identification and encrypted storage, specifically by obtaining the user information verification feedback information, combining the identification summary in the data identification data information, and performing big data management optimization based on the storage feedback information to obtain data storage management feedback information.
6. A big data management optimization system for data storage, used to implement a big data management optimization method for data storage as described in any one of claims 1 to 5, Features: It includes a data system building module, a user information verification module, an input data identification module, a data encryption storage module and a big data management optimization module.
7. A big data management optimization system for data storage according to claim 6, Features: The data system construction module is used for data system construction. Through the data system construction, a data management basic subsystem is obtained, and the data management basic subsystem is used for user information verification, input data identification and data encryption storage; The user information verification module is used for user information verification, obtains user information verification feedback information and data to be stored through user information verification, sends the data to be stored to the input data identification module, and sends the user information verification feedback information to the big data management optimization module; The input data identification module is used for input data identification, obtains data identification data information through input data identification, and sends the data identification data information to the data encryption storage module and the big data management optimization module; The data encryption storage module is used for data encryption storage, obtains storage feedback information through data encryption storage, and sends the storage feedback information to the big data management optimization module; The big data management optimization module is used for big data management optimization, and data storage management feedback information is obtained through big data management optimization.
Citation Information
Patent Citations
Double-fed fan control parameter identification method based on LSTM-IPSO
CN116227320A
