A Deep Learning-Based Online Data Service Center for E-readers

Abstract

The present invention discloses an electronic reader online data service center based on deep learning, which belongs to the field of data identification technology. When data is interacted with an electronic reader, network flow information generated in the interaction process is collected, and a network flow identification model with network security monitoring is used to identify the network flow information, so that anomalies in the interaction process between the electronic reader and the data server center can be discovered, illegal attack behaviors can be discovered in time, and the security performance of the electronic reader online data service center is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data recognition, and in particular relates to an online data service center for electronic readers based on deep learning. Background Art

[0002] The e-reader online data service center is a platform that uses modern technology to provide digital reading services. This service provides users with an integrated digital reading experience including book search, book reading, book listening, and book purchase through digital technologies such as cloud computing and big data. The cloud reading platform not only aggregates a large number of content resources such as books, journals, and web articles, but also supports readers to conduct high-quality digital reading on different reading terminals. These online data service centers often use conventional firewalls to ensure service security, but traditional firewall technology can no longer meet the security needs of data services due to its lack of ability to respond to new network attacks. Summary of the invention

[0003] The present invention provides an electronic reader online data service center based on deep learning, which is used to solve the technical problems existing in the prior art.

[0004] An online data service center for electronic readers based on deep learning, comprising: a data storage module, a network traffic collection module, a deep learning module and a security supervision module;

[0005] The data storage module is used to store online data of the electronic reader and allow the electronic reader to access the stored electronic book data;

[0006] The network traffic collection module is used to collect network traffic information generated during data interaction with the electronic reader;

[0007] The deep learning module is used to construct a network traffic identification model for security supervision using a deep learning model, and after training the network traffic identification model, obtain a network traffic identification model with network security monitoring capabilities;

[0008] The security supervision module is used to identify network traffic information using a network traffic identification model with network security monitoring, obtain network security supervision results, and control access to the electronic reader based on the network security supervision results.

[0009] Furthermore, a deep learning model is used to construct a network traffic identification model for security supervision, including: a network intrusion detection model is constructed by sequentially connecting a first convolutional layer, a first ReLU activation layer, a first pooling layer, a second convolutional layer, a second ReLU activation layer, a second pooling layer, a third convolutional layer, a third ReLU activation layer, a third pooling layer, a Flatten layer, a fully connected layer, a Dropout layer and a Softmax output layer.

[0010] Furthermore, after training the network traffic identification model, a network traffic identification model with network security monitoring is obtained, including:

[0011] Initializing the hyperparameters of the network traffic identification model multiple times to obtain hyperparameter encoding;

[0012] After repeated initialization, multiple different hyperparameter encodings are obtained;

[0013] Obtain the fitness corresponding to each hyperparameter encoding, and determine the optimal hyperparameter encoding according to the fitness corresponding to each hyperparameter encoding;

[0014] Taking the optimal hyperparameter encoding as a reference, a nonlinear weighted bootstrapping strategy is used to perform a bootstrapping update on each hyperparameter encoding to obtain a hyperparameter encoding after the bootstrapping update;

[0015] A double optimal bootstrap strategy is used to perform a secondary bootstrap update on the hyperparameter encoding after the primary bootstrap update, and the hyperparameter encoding after the secondary bootstrap update is obtained;

[0016] A random matching cross-guided update strategy is used to perform a three-step guide update on the hyperparameter encoding after the two-step guide update to obtain the hyperparameter encoding after the three-step guide update;

[0017] The random flight guidance strategy is used to guide the hyperparameter encoding after three guide updates for four times, and the hyperparameter encoding after four guide updates is obtained;

[0018] Determine whether the iteration end condition is met. If so, re-obtain the optimal hyperparameter encoding, and use the re-obtained hyperparameter encoding as the final parameter of the network traffic identification model to obtain a network traffic identification model with network security monitoring. Otherwise, return to the step of guiding the update once.

[0019] Furthermore, the fitness corresponding to each hyperparameter encoding is obtained, including:

[0020] Acquire a training set that is pre-stored or input through human-computer interaction; wherein the training set includes sample network traffic information and sample labels corresponding to the sample network traffic information;

[0021] After applying the hyperparameters included in the hyperparameter encoding to the network traffic identification model, using the sample network traffic information as the input of the network traffic identification model to obtain the actual output of the network traffic identification model;

[0022] Determine the cross entropy loss function corresponding to the hyperparameter encoding according to the actual output of the network traffic recognition model and the sample labels corresponding to the sample network traffic information;

[0023] According to the cross entropy loss function corresponding to the hyperparameter encoding, the fitness corresponding to the hyperparameter encoding is determined to be: 1 / (cross entropy loss function + 0.0001);

[0024] Traverse all hyperparameter encodings and get the fitness corresponding to each hyperparameter encoding.

[0025] Furthermore, determining the optimal hyperparameter encoding according to the fitness corresponding to each hyperparameter encoding includes: determining the hyperparameter encoding with the smallest fitness as the optimal hyperparameter encoding according to the fitness corresponding to each hyperparameter encoding.

[0026] Further, taking the optimal hyperparameter encoding as a reference, a nonlinear weighted guidance strategy is used to perform a guidance update on each hyperparameter encoding to obtain a hyperparameter encoding after a guidance update, including:

[0027] For any hyperparameter encoding, weighted subtraction is performed on the optimal hyperparameter encoding and the hyperparameter encoding to determine the guidance information item;

[0028] combining the guidance information items using a first nonlinear search coefficient to obtain first guidance update information;

[0029] The hyperparameter encoding is weighted using a second nonlinear search coefficient to obtain the original position information;

[0030] The original position information is merged with the first guided update information to obtain an updated hyperparameter encoding.

[0031] Furthermore, a dual optimal guidance strategy is used to perform a secondary guidance update on the hyperparameter encoding after the primary guidance update, and the hyperparameter encoding after the secondary guidance update is obtained, including:

[0032] For any hyperparameter code after a guided update, determine a first guided update coefficient and a second guided update coefficient according to the fitness corresponding to each hyperparameter code;

[0033] Determining a second guided update item corresponding to the hyperparameter encoding according to the optimal hyperparameter encoding, the historical optimal state corresponding to the hyperparameter encoding, and the first guided update coefficient;

[0034] Determining a third guided update item corresponding to the hyperparameter encoding according to the optimal hyperparameter encoding, the historical optimal state corresponding to the hyperparameter encoding, and the second guided update coefficient;

[0035] According to the second boot update item and the third boot update item, the hyperparameter code is boot-updated for a second time to obtain the hyperparameter code after the boot update for the second time.

[0036] Furthermore, a random matching cross-guided update strategy is used to perform a three-times guided update on the hyperparameter encoding after the two-times guided update, and the hyperparameter encoding after the three-times guided update is obtained, including:

[0037] For any hyperparameter code after the secondary bootstrap update, randomly match a first random code and a second random code for the hyperparameter code;

[0038] According to the first random code and the second random code, an adaptive search control coefficient is used to obtain a fourth guided update item corresponding to the hyperparameter code;

[0039] According to the fourth guided update item corresponding to the hyperparameter code, the hyperparameter code is guided and updated three times to obtain the hyperparameter code after the three guided updates.

[0040] Furthermore, the random flight guidance strategy is used to perform four guidance updates on the hyperparameter encoding after the three guidance updates, and the hyperparameter encoding after the four guidance updates is obtained, including:

[0041] For each hyperparameter encoding after three bootstrap updates, the optimal hyperparameter encoding is used as a reference to determine the information difference term corresponding to the hyperparameter encoding;

[0042] The fourth guided update term is obtained by weighting the information difference term corresponding to the hyperparameter encoding using random flight;

[0043] According to the fourth guided update item, the hyperparameter encoding is guided and updated four times to obtain the hyperparameter encoding after the four guided updates.

[0044] Furthermore, a network traffic identification model with network security monitoring is used to identify network traffic information, obtain network security supervision results, and control access to the electronic reader according to the network security supervision results, including:

[0045] Using the network traffic information as input to a network traffic identification model with network security monitoring to obtain network security supervision results; wherein the network security supervision results include network security or network anomalies;

[0046] When the network security supervision result is network security, maintaining the access connection with the electronic reader;

[0047] When the network security supervision result is a network anomaly, the access connection with the electronic reader is disconnected, and a network security anomaly log is generated based on the information of the electronic reader.

[0048] The present invention provides an electronic reader online data service center based on deep learning. When interacting with the electronic reader for data, the center collects network traffic information generated during the interaction process, and uses a network traffic identification model with network security monitoring to identify the network traffic information. Thus, anomalies in the interaction process between the electronic reader and the data server center can be discovered, illegal attack behaviors can be discovered in a timely manner, and the security performance of the electronic reader online data service center is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0050] Figure 1 A schematic diagram of the structure of an online data service center for an electronic reader based on deep learning provided in an embodiment of the present invention.

[0051] Figure 2 A flowchart for training the network traffic identification model provided in an embodiment of the present invention.

[0052] Among them, 101-data storage module, 102-network traffic collection module, 103-deep learning module, and 104-security supervision module.

[0053] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0054] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0055] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0056] like Figure 1As shown, an embodiment of the present invention provides an electronic reader online data service center based on deep learning, including: a data storage module 101, a network traffic collection module 102, a deep learning module 103 and a security supervision module 104;

[0057] The data storage module 101 is used to store online data of the electronic reader and allow the electronic reader to access the stored electronic book data;

[0058] The data storage module 101 is mainly used to provide online data services for the electronic reader, which is consistent with the prior art and will not be described in detail in the embodiment of the present invention.

[0059] The network traffic collection module 102 is used to collect network traffic information generated during data interaction with the electronic reader;

[0060] Network traffic information is used to reflect the characteristics of network security during data interaction. By collecting and identifying network traffic information, the security of the e-reader's online data service center can be effectively tested.

[0061] The deep learning module 103 is used to construct a network traffic identification model for security supervision using a deep learning model, and after training the network traffic identification model, obtain a network traffic identification model with network security monitoring capabilities;

[0062] A network traffic identification model can be constructed by using deep learning models such as convolutional neural networks and BP (Back Propagation) neural networks. Then, the hyperparameters of the network traffic identification model can be optimized by using methods such as gradient descent and intelligent optimization algorithms, thereby obtaining a network traffic identification model with network security monitoring capabilities.

[0063] The security supervision module 104 is used to identify network traffic information using a network traffic identification model with network security monitoring, obtain network security supervision results, and control access to the electronic reader according to the network security supervision results.

[0064] Controlling access to the electronic book reader according to the network security supervision result may include: generating warning information and / or interrupting the connection with the electronic book reader when there is an abnormality in the network security supervision result.

[0065] The present invention provides an electronic reader online data service center based on deep learning. When interacting with the electronic reader for data, the center collects network traffic information generated during the interaction process, and uses a network traffic identification model with network security monitoring to identify the network traffic information. Thus, anomalies in the interaction process between the electronic reader and the data server center can be discovered, illegal attack behaviors can be discovered in a timely manner, and the security performance of the electronic reader online data service center is improved.

[0066] In an embodiment of the present invention, a deep learning model is used to construct a network traffic identification model for security supervision, including: using a first convolutional layer, a first ReLU activation layer, a first pooling layer, a second convolutional layer, a second ReLU activation layer, a second pooling layer, a third convolutional layer, a third ReLU activation layer, a third pooling layer, a Flatten layer, a fully connected layer, a Dropout layer (regularization layer) and a Softmax output layer connected in sequence to construct a network intrusion detection model.

[0067] It is worth noting that the above network traffic identification model is only a preferred implementation of the present invention, and other neural network models can also be used to construct a network traffic identification model, for example, a convolutional neural network or a BP neural network can be used to construct a network traffic identification model.

[0068] After building a network traffic identification model, it is usually necessary to optimize the hyperparameters of the network traffic identification model, generally optimizing the connection weights and thresholds between the layers, so that the network traffic identification model can accurately identify network traffic information and ultimately identify whether the online data service center of the e-reader is safe. However, in the prior art, an algorithm such as a particle swarm algorithm is often used for hyperparameter optimization, but it usually falls into a local optimum, and the convergence speed and convergence effect are not ideal. Therefore, an embodiment of the present invention provides a method for training the network traffic identification model, specifically referring to the following method.

[0069] like Figure 2 As shown, after the network traffic identification model is trained, a network traffic identification model with network security monitoring is obtained, including:

[0070] S201, initializing the hyperparameters of the network traffic identification model multiple times to obtain hyperparameter codes;

[0071] For example, the hyperparameters of a network traffic identification model have upper and lower limits, and can usually be randomly initialized between the upper and lower limits of the hyperparameters to obtain hyperparameter encoding.

[0072] It is worth noting that, in the embodiment of the present invention, the initialization may be all the hyperparameters of the network traffic identification model, or may be part of the hyperparameters of the network traffic identification model.

[0073] S202, after repeated initialization, multiple different hyperparameter encodings are obtained;

[0074] S203, obtaining the fitness corresponding to each hyperparameter encoding, and determining the optimal hyperparameter encoding according to the fitness corresponding to each hyperparameter encoding;

[0075] The fitness is used to reflect the quality of the position of the hyperparameter encoding. Therefore, the larger the fitness, the better the position of the hyperparameter encoding in the solution space, so that the optimal hyperparameter encoding can be determined.

[0076] S204, taking the optimal hyperparameter code as a reference, adopting a nonlinear weighted guidance strategy to perform a guidance update on each hyperparameter code, to obtain a hyperparameter code after the guidance update;

[0077] S205, performing a secondary guidance update on the hyperparameter code after the primary guidance update using a dual optimal guidance strategy to obtain the hyperparameter code after the secondary guidance update;

[0078] S206, performing a three-times bootstrap update on the hyperparameter code after the two-times bootstrap update by adopting a random matching cross-bootstrap update strategy, to obtain the three-times bootstrap update hyperparameter code;

[0079] S207, using a random flight guidance strategy to perform four guidance updates on the hyperparameter code after three guidance updates, to obtain the hyperparameter code after four guidance updates;

[0080] S208. Determine whether the iteration end condition is met. If so, re-obtain the optimal hyperparameter encoding, and use the re-obtained hyperparameter encoding as the final parameter of the network traffic identification model to obtain a network traffic identification model with network security monitoring. Otherwise, return to the step of guiding the update once.

[0081] The iteration end condition can be set as: the current training number is greater than the maximum training number or the maximum fitness is greater than a preset threshold.

[0082] In the embodiment of the present invention, obtaining the fitness corresponding to each hyperparameter encoding includes:

[0083] Acquire a training set that is pre-stored or input through human-computer interaction; wherein the training set includes sample network traffic information and sample labels corresponding to the sample network traffic information;

[0084] After applying the hyperparameters included in the hyperparameter encoding to the network traffic identification model, using the sample network traffic information as the input of the network traffic identification model to obtain the actual output of the network traffic identification model;

[0085] Determine the cross entropy loss function corresponding to the hyperparameter encoding according to the actual output of the network traffic recognition model and the sample labels corresponding to the sample network traffic information;

[0086] According to the cross entropy loss function corresponding to the hyperparameter encoding, the fitness corresponding to the hyperparameter encoding is determined to be: 1 / (cross entropy loss function + 0.0001);

[0087] Traverse all hyperparameter encodings and get the fitness corresponding to each hyperparameter encoding.

[0088] The training set can use the network intrusion detection dataset KDD Cup 1999, which has nearly 5 million data records. Each record contains 42 columns. The first 41 columns are feature attributes, and the last column is a label, which indicates which type of attack this row of data belongs to.

[0089] Since the dimension of feature attributes is very high, according to previous experience in machine learning, there must be irrelevant and redundant features. If these features are not removed, on the one hand, the accuracy of model detection will decrease, and on the other hand, when encoding and decoding the feature attributes of the data, the complexity of calculation will increase, resulting in a decrease in the convergence speed of the model, an increase in training time, and more memory usage, and the requirements for computer performance will become more stringent. Therefore, it is also possible to perform dimensionality reduction processing on sample network traffic information and network traffic information.

[0090] In an embodiment of the present invention, the optimal hyperparameter encoding is determined according to the fitness corresponding to each hyperparameter encoding, including: according to the fitness corresponding to each hyperparameter encoding, determining the hyperparameter encoding with the smallest fitness as the optimal hyperparameter encoding.

[0091] In the embodiment of the present invention, taking the optimal hyperparameter encoding as a reference, a nonlinear weighted guiding strategy is used to perform a guided update on each hyperparameter encoding to obtain a hyperparameter encoding after the guided update, including:

[0092] For any hyperparameter encoding, weighted subtraction is performed on the optimal hyperparameter encoding and the hyperparameter encoding to determine the guided information item as ( );in, represents the first weighting coefficient, and , represents pi, represents the second weighting coefficient, and , Indicates tDuring the training m Hyperparameter encoding, m= 1,2,…,M, where M represents the total number of hyperparameter encodings. represents the optimal hyperparameter encoding;

[0093] Using the first nonlinear search coefficient The boot information items are combined to obtain the first boot update information: * ;in, represents the first random number between [0,2π], represents the second random number between [0,π];

[0094] Using the second nonlinear search coefficient | |Weight the hyperparameter encoding and get the original position information as | |* ;

[0095] The original position information is fused with the first guided update information to obtain an updated hyperparameter encoding: =| |* + * ;in, Represents the hyperparameter encoding after an update .

[0096] The nonlinear weighted guidance strategy provided in the embodiment of the present invention can be based on the optimal hyperparameter encoding, and adopts nonlinear moving speed and irregular moving rules to move in the optimal direction, that is, it can realize the search of local areas and can also search more unknown areas, thereby effectively improving the algorithm convergence speed.

[0097] In an embodiment of the present invention, a dual optimal guidance strategy is used to perform a secondary guidance update on the hyperparameter encoding after the primary guidance update to obtain the hyperparameter encoding after the secondary guidance update, including:

[0098] For any hyperparameter code after a guided update, determine a first guided update coefficient and a second guided update coefficient according to the fitness corresponding to each hyperparameter code;

[0099] Among them, the first guided update coefficient is:

[0100]

[0101] The second guided update coefficient is:

[0102]

[0103] in, represents the first guided update coefficient during the t-th training process, represents the maximum fitness corresponding to the hyperparameter encoding after a guided update, represents the minimum fitness corresponding to the hyperparameter encoding after a guided update, represents the average fitness corresponding to the hyperparameter encoding after a guided update, represents a constant factor between (0,1), represents the second guided update coefficient during the t-th training process.

[0104] The embodiment of the present invention can adaptively adjust the search speed by setting the first guided update coefficient and the second guided update coefficient with adaptive changes, thereby effectively improving the convergence speed of the algorithm.

[0105] According to the optimal hyperparameter encoding, the historical optimal state corresponding to the hyperparameter encoding, and the first guided update coefficient, the second guided update item corresponding to the hyperparameter encoding is determined as: ;in, represents the optimal hyperparameter encoding, Indicates n The historical optimal state of the hyperparameter encoding after a guided update, Indicates n The hyperparameter encoding after a bootstrap update, m= 1,2,…,M;

[0106] According to the optimal hyperparameter encoding, the historical optimal state corresponding to the hyperparameter encoding, and the second guided update coefficient, the third guided update item corresponding to the hyperparameter encoding is determined as: ;in, Indicates that when the fitness is greater than the hyperparameter encoding Randomly select paired hyperparameter encodings from the encodings of ;

[0107] According to the second guided update item and the third guided update item, the hyperparameter code is guided twice to update, and the hyperparameter code after the secondary guided update is obtained as follows: ;in, Represents the hyperparameter encoding after the second bootstrap update .

[0108] The dual optimal guidance strategy provided by the embodiment of the present invention can enable sufficient information interaction and effectively prevent the algorithm from falling into a local optimum.

[0109] In the embodiment of the present invention, a random matching cross-guided update strategy is used to perform a three-guided update on the hyperparameter code after the two-guided update, and the hyperparameter code after the three-guided update is obtained, including:

[0110] For any hyperparameter code after the second bootstrap update, randomly match the hyperparameter code with a first random code And the second random code ;

[0111] According to the first random coding And the second random code , the fourth guided update term corresponding to the hyperparameter encoding is obtained by adaptively searching the control coefficients: ;in, s represents the boot update step size, represents the adaptive search control coefficient, and , T Indicates the maximum number of training times;

[0112] According to the fourth guided update item corresponding to the hyperparameter code, the hyperparameter code is guided and updated three times, and the hyperparameter code after the three guided updates is obtained as follows: ; Indicates k The hyperparameter encoding after the second bootstrap update, Represents the hyperparameter encoding after three bootstrap updates , k= 1,2,…,M;

[0113] The random matching cross-guidance update strategy of the embodiment of the present invention can perform a larger range of searches under the premise of the algorithm, and as the algorithm proceeds, the convergence accuracy can be effectively improved, thereby improving the convergence effect of the algorithm.

[0114] In the embodiment of the present invention, a random flight guidance strategy is used to perform four guidance updates on the hyperparameter coding after three guidance updates, and the hyperparameter coding after four guidance updates is obtained, including:

[0115] For each hyperparameter encoding after three bootstrap updates, the optimal hyperparameter encoding is used as a reference to determine the information difference term corresponding to the hyperparameter encoding ;in, represents the Levy flight code;

[0116] The information difference term corresponding to the hyperparameter encoding is weighted by random flight to obtain the fourth guided update term * ;in, l Represents a random number between [-1,1], Indicates the upper limit With lower limit The random coefficients between π represents pi;

[0117] According to the fourth guided update item, the hyperparameter encoding is guided and updated four times, and the hyperparameter encoding after the four guided updates is obtained as follows: = + * ;in, Indicates i The hyperparameter encoding after three bootstrap updates, Represents the hyperparameter encoding after four bootstrap updates , i= 1,2,…,M.

[0118] Although the above algorithm has a certain global search capability, the global search capability is limited. In order to prevent the algorithm from falling into the local optimum, the embodiment of the present invention adopts a random flight guidance strategy to search in the solution space in an irregular flying manner, thereby preventing the algorithm from falling into the local optimum.

[0119] Optionally, in the stage involving global search, the embodiment of the present invention may also use a greedy strategy to control the global search, thereby ensuring the convergence speed of the algorithm. It may also perform out-of-bounds processing after each search to ensure the validity of the hyperparameters.

[0120] In an embodiment of the present invention, a network traffic identification model with network security monitoring is used to identify network traffic information, obtain network security supervision results, and control access to the electronic reader according to the network security supervision results, including:

[0121] Using the network traffic information as input to a network traffic identification model with network security monitoring to obtain network security supervision results; wherein the network security supervision results include network security or network anomalies;

[0122] When the network security supervision result is network security, maintaining the access connection with the electronic reader;

[0123] When the network security supervision result is a network anomaly, the access connection with the electronic reader is disconnected, and a network security anomaly log is generated based on the information of the electronic reader.

[0124] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and interpreted scripting language JavaScript, etc.

[0125] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0128] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0129] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An electronic reader online data service center based on deep learning, characterized in that: include: Data storage module, network traffic collection module, deep learning module and security supervision module; The data storage module is used to store online data of the electronic reader and allow the electronic reader to access the stored electronic book data; The network traffic collection module is used to collect network traffic information generated during data interaction with the electronic reader; The deep learning module is used to construct a network traffic identification model for security supervision using a deep learning model, and after training the network traffic identification model, obtain a network traffic identification model with network security monitoring capabilities; The security supervision module is used to identify network traffic information using a network traffic identification model with network security monitoring, obtain network security supervision results, and control access to the electronic reader according to the network security supervision results; A deep learning model is used to build a network traffic identification model for security supervision, including: a network intrusion detection model is built by sequentially connecting the first convolution layer, the first ReLU activation layer, the first pooling layer, the second convolution layer, the second ReLU activation layer, the second pooling layer, the third convolution layer, the third ReLU activation layer, the third pooling layer, the Flatten layer, the fully connected layer, the Dropout layer and the Softmax output layer; After training the network traffic identification model, a network traffic identification model with network security monitoring is obtained, including: Initializing the hyperparameters of the network traffic identification model multiple times to obtain hyperparameter encoding; After repeated initialization, multiple different hyperparameter encodings are obtained; Obtain the fitness corresponding to each hyperparameter encoding, and determine the optimal hyperparameter encoding according to the fitness corresponding to each hyperparameter encoding; Taking the optimal hyperparameter encoding as a reference, a nonlinear weighted bootstrapping strategy is used to perform a bootstrapping update on each hyperparameter encoding to obtain a hyperparameter encoding after the bootstrapping update; A double optimal bootstrap strategy is used to perform a secondary bootstrap update on the hyperparameter encoding after the primary bootstrap update, and the hyperparameter encoding after the secondary bootstrap update is obtained; A random matching cross-guided update strategy is used to perform a three-step guide update on the hyperparameter encoding after the two-step guide update to obtain the hyperparameter encoding after the three-step guide update; The random flight guidance strategy is used to guide the hyperparameter encoding after three guide updates for four times, and the hyperparameter encoding after four guide updates is obtained; Determine whether the iteration end condition is met. If so, re-obtain the optimal hyperparameter encoding, and use the re-obtained hyperparameter encoding as the final parameter of the network traffic identification model to obtain a network traffic identification model with network security monitoring. Otherwise, return to the step of guiding the update once.

2. The electronic reader online data service center based on deep learning according to claim 1, characterized in that: Get the fitness corresponding to each hyperparameter encoding, including: Acquire a training set that is pre-stored or input through human-computer interaction; wherein the training set includes sample network traffic information and sample labels corresponding to the sample network traffic information; After applying the hyperparameters included in the hyperparameter encoding to the network traffic identification model, using the sample network traffic information as the input of the network traffic identification model to obtain the actual output of the network traffic identification model; Determine the cross entropy loss function corresponding to the hyperparameter encoding according to the actual output of the network traffic recognition model and the sample labels corresponding to the sample network traffic information; According to the cross entropy loss function corresponding to the hyperparameter encoding, the fitness corresponding to the hyperparameter encoding is determined to be: 1 / (cross entropy loss function + 0.0001); Traverse all hyperparameter encodings and get the fitness corresponding to each hyperparameter encoding.

3. The deep learning-based online data service center for electronic readers according to claim 1 or 2, characterized in that: Determining the optimal hyperparameter encoding according to the fitness corresponding to each hyperparameter encoding includes: determining the hyperparameter encoding with the smallest fitness as the optimal hyperparameter encoding according to the fitness corresponding to each hyperparameter encoding.

4. The electronic reader online data service center based on deep learning according to claim 1, characterized in that: Taking the optimal hyperparameter encoding as a reference, a nonlinear weighted bootstrap strategy is used to perform a bootstrap update on each hyperparameter encoding to obtain a hyperparameter encoding after a bootstrap update, including: For any hyperparameter encoding, weighted subtraction is performed on the optimal hyperparameter encoding and the hyperparameter encoding to determine the guidance information item; combining the guidance information items using a first nonlinear search coefficient to obtain first guidance update information; The hyperparameter encoding is weighted using a second nonlinear search coefficient to obtain the original position information; The original position information is merged with the first guided update information to obtain an updated hyperparameter encoding.

5. The deep learning-based online data service center for electronic readers according to claim 4, characterized in that: The double optimal bootstrap strategy is used to perform a secondary bootstrap update on the hyperparameter encoding after the primary bootstrap update, and the hyperparameter encoding after the secondary bootstrap update is obtained, including: For any hyperparameter code after a guided update, determine a first guided update coefficient and a second guided update coefficient according to the fitness corresponding to each hyperparameter code; Determining a second guided update item corresponding to the hyperparameter encoding according to the optimal hyperparameter encoding, the historical optimal state corresponding to the hyperparameter encoding, and the first guided update coefficient; Determining a third guided update item corresponding to the hyperparameter encoding according to the optimal hyperparameter encoding, the historical optimal state corresponding to the hyperparameter encoding, and the second guided update coefficient; According to the second boot update item and the third boot update item, the hyperparameter code is boot-updated for a second time to obtain the hyperparameter code after the boot update for the second time.

6. The electronic reader online data service center based on deep learning according to claim 5, characterized in that: The random matching cross-guided update strategy is used to perform three-times-guided update on the hyperparameter encoding after the two-times-guided update, and the hyperparameter encoding after the three-times-guided update is obtained, including: For any hyperparameter code after the secondary bootstrap update, randomly match a first random code and a second random code for the hyperparameter code; According to the first random code and the second random code, an adaptive search control coefficient is used to obtain a fourth guided update item corresponding to the hyperparameter code; According to the fourth guided update item corresponding to the hyperparameter code, the hyperparameter code is guided and updated three times to obtain the hyperparameter code after the three guided updates.

7. The deep learning-based online data service center for electronic readers according to claim 6, characterized in that: The random flight guidance strategy is used to perform four guidance updates on the hyperparameter encoding after three guidance updates, and the hyperparameter encoding after four guidance updates is obtained, including: For each hyperparameter encoding after three bootstrap updates, the optimal hyperparameter encoding is used as a reference to determine the information difference term corresponding to the hyperparameter encoding; The fourth guided update term is obtained by weighting the information difference term corresponding to the hyperparameter encoding using random flight; According to the fourth guided update item, the hyperparameter encoding is guided and updated four times to obtain the hyperparameter encoding after the four guided updates.

8. The deep learning-based online data service center for electronic readers according to claim 7, characterized in that: The network traffic information is identified by using a network traffic identification model with network security monitoring to obtain a network security supervision result, and the access to the electronic reader is controlled according to the network security supervision result, including: Using the network traffic information as input to a network traffic identification model with network security monitoring to obtain network security supervision results; wherein the network security supervision results include network security or network anomalies; When the network security supervision result is network security, maintaining the access connection with the electronic reader; When the network security supervision result is a network anomaly, the access connection with the electronic reader is disconnected, and a network security anomaly log is generated based on the information of the electronic reader.

Images