Intelligent control method for data unidirectional transmission system based on LSTM prediction analysis

Abstract

The present application relates to the field of communication technology, especially to a kind of data unidirectional transmission system intelligent control method based on LSTM prediction analysis, first, the training set containing time series data is sorted out;Then construct the neural network architecture based on LSTM, input the training set into neural network architecture, train neural network architecture;Neural network architecture is adjusted during training, until the best model is obtained after training ends;Finally, the best model obtained is deployed to actual data unidirectional transmission system, and the transmission delay time of data packet is predicted according to the best model in unidirectional transmission system, and the transmission delay time is adjusted and sent according to the predicted transmission delay time.The present application realizes the parameter self-adaptive prediction and auxiliary adjustment of unidirectional data transmission system by constructing the neural network architecture based on LSTM, improves the transmission success rate, reduces manual intervention and system overhead, and enhances the robustness of the system.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data communication transmission, and particularly relates to an intelligent control method for a data one-way transmission system based on LSTM prediction analysis. BACKGROUND

[0002] One-way data transmission (also known as data diode technology) is an important data security technology. Its essence is to use hardware to ensure that data can only be transmitted from a low-security level network to a high-security level network, and sensitive data will not be leaked due to reverse transmission. This transmission method is usually used in highly secure scenarios (such as military, government agencies, financial institutions, etc.) to protect sensitive data from unauthorized access and leakage.

[0003] The one-way data transmission system cannot form a retransmission mechanism, so the traffic of the transmission process is controlled by the sending end. In a specific performance computer and network channel, the data packet size and sending delay of the sending end determine the data transmission rate and success rate. The size of the network data packet or the sending delay is proportional to the transmission efficiency and inversely proportional to the transmission success rate, that is, larger data packets or shorter sending delays can improve transmission efficiency, but may increase the risk of data loss and reduce the success rate of data transmission; smaller data packets or longer sending delays are more suitable for unstable network environments and help improve the success rate of data transmission, but limit the data transmission rate. Suitable data packets and sending delays can prevent network congestion and buffer overflow, and the transmission rate needs to be adjusted according to the network performance and the processing capacity of the receiving end.

[0004] Traditional control strategies include manual control, delayed redundant sending, and dual-machine redundant feedback methods. Among them:

[0005] (1) Manual control, that is, if the receiving end data cannot be successfully parsed or part of the data does not match during transmission, it means that the transmission is incorrect, and the receiving end needs to manually report the error. The staff then informs the sender to resend the message through other means. Although it ensures the reliability of transmission, it greatly consumes manpower and resources, and the efficiency is low;

[0006] (2) Delayed redundant sending, that is, the sending end increases the redundancy of the data packet and transmits the data according to a certain delay period and repetition number. This method improves the fault tolerance of one-way data transmission, but high redundancy leads to low efficiency, especially when the data size is large, which has a great impact on business;

[0007] (3) Dual-machine redundant feedback, that is, two sets of one-way transmission systems with the same configuration are built to realize the feedback mechanism by comparing the sending and receiving of the two systems, which can alleviate the above problems to a certain extent, but increases the hardware cost. SUMMARY

[0008] In view of this, the present application aims to provide a data unidirectional transmission system intelligent control method based on LSTM prediction analysis, by constructing a neural network architecture based on LSTM (Long Short-Term Memory), realizing the parameter adaptive prediction and auxiliary adjustment of the unidirectional data transmission system, so as to find the best balance point between the data transmission rate and the data transmission success rate, improve the transmission success rate, reduce manual intervention and system overhead, and enhance the system robustness.

[0009] To achieve the above-mentioned purpose, the technical scheme of the present application is realized as follows:

[0010] A data unidirectional transmission system intelligent control method based on LSTM prediction analysis, comprising the following steps:

[0011] S1: Collecting time series data affecting the data transmission rate and the data transmission success rate, and pre-processing the time series data to obtain a training set;

[0012] S2: Constructing a neural network architecture based on LSTM, inputting the training set obtained in step S1 into the neural network architecture, training the neural network architecture, and obtaining a network model;

[0013] S3: Inputting the time series data to be tested into the network model obtained in step S2 for prediction, adjusting the neural network architecture according to the prediction result; repeating step S2 to train the adjusted neural network architecture until the training is completed to obtain an optimal model;

[0014] S4: Deploying the optimal model obtained in step S3 to an actual data unidirectional transmission system, and the unidirectional transmission system predicting the transmission delay time of the data packet according to the optimal model, and adjusting and sending the data packet according to the predicted transmission delay time.

[0015] Further, in step S1, the time series data includes network I / O performance data, hardware I / O performance data and data packet size of the sending end.

[0016] Further, the pre-processing in step S1 comprises the following steps:

[0017] S11: Normalizing the time series data;

[0018] S12: Organizing the normalized time series data into a sequence according to the time , and further creating a feature sequence for:

[0019] ;

[0020] The corresponding data packet average delay time , express The average packet delay time at time, t Indicates the moment, T Indicates the total time;

[0021] Feature Sequence Each element in Contains K features, namely:

[0022] ;

[0023] S13: The feature sequence Further divide according to the time step n to obtain the training feature sequence :

[0024] ;

[0025] ;

[0026] in, m Represents the training feature sequence Neutron sequence The index of Represents the training feature sequence Neutron sequence The total number of each subsequence ;

[0027] S14: Training feature sequence and the corresponding average packet delay Construct a training set and train feature sequence subsequence in correspond Average packet delay at time .

[0028] Furthermore, in step S2, the neural network architecture includes an input layer, a hidden layer, and an output layer; wherein the input of the hidden layer at the current moment includes the subsequence corresponding to the current moment , the output of the hidden layer at the previous moment and the state information of the hidden layer at the previous moment; the output of the hidden layer at the current moment includes the output of the hidden layer at the current moment and the state information of the hidden layer at the current moment; the output result of the hidden layer at the last moment passes through the fully connected layer and outputs the predicted average delay time of the data packet through the output layer.

[0029] Further, the hidden layer comprises an input gate, a forgetting gate, an output gate and a change gate; the input gate determines information input into the hidden layer at a current time, the forgetting gate determines a forgetting degree of information of the hidden layer at a previous time by the hidden layer at the current time, the change gate determines state information of the hidden layer at the current time according to information of the forgetting gate and the input gate, and the output gate outputs output of the hidden layer at the current time according to the state information of the hidden layer at the current time.

[0030] Further, in step S4, the one-way transmission system comprises a first thread and a second thread, the second thread is used to predict a transmission delay time of the data packet by the best model, and the predicted transmission delay time is transmitted to the first thread; the first thread is used to package the data, and the data packet is adjusted and sent according to the predicted transmission delay time.

[0031] Compared with the prior art, the application can achieve the following beneficial effects:

[0032] (1) In the data one-way transmission system intelligent control method based on LSTM prediction analysis, the transmission strategy can be dynamically adjusted according to network congestion, delay and other conditions through the accurate prediction of the LSTM model, so as to avoid packet loss or transmission failure, significantly improve the success rate of data transmission, and ensure the safe and reliable transmission of data; at the same time, the application can collect and analyze network and hardware performance data in real time, predict the data transmission success rate, dynamically adjust the transmission parameters (such as data packet size, sending delay, etc.), and realize the optimal balance between transmission rate and success rate;

[0033] (2) The data one-way transmission system intelligent control method based on LSTM prediction analysis reduces manual intervention and improves efficiency: the traditional method relies on manual intervention (such as manual reporting and resending data), which not only consumes manpower, but also easily leads to operation delay; the application realizes automatic prediction and adjustment through the LSTM model, greatly reduces manual participation, improves the automation level of the system and the data transmission efficiency, and is especially suitable for scenarios that require long-time and efficient transmission;

[0034] (3) The data one-way transmission system intelligent control method based on LSTM prediction analysis supports large-scale data transmission: in the process of large-scale data transmission, the traditional redundant sending mechanism often significantly reduces the transmission efficiency; the application can maintain high efficiency and high reliability in large data transmission by optimizing the transmission parameters and rate, and is more suitable for large-scale data one-way transmission requirements in the fields of industry, government and finance. BRIEF DESCRIPTION OF DRAWINGS

[0035] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The illustrations, together with their description, serve to explain the application without limiting the application thereto.

[0036] Figure 1 A flowchart of the intelligent control method of the data unidirectional transmission system based on LSTM prediction analysis according to an embodiment of the present application;

[0037] Figure 2 A structural diagram of the neural network architecture based on LSTM according to an embodiment of the present application;

[0038] Figure 3 A structural diagram of the hidden layer according to an embodiment of the present application;

[0039] Figure 4 A flowchart of step S4 according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and do not constitute a limitation on the present application.

[0041] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0042] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the present application and simplify the description, and therefore cannot be understood as indicating or implying that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second" and the like are only used for description purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features limited by "first", "second" and the like can explicitly or implicitly include one or more features. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0043] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "linking" should be understood in a broad sense, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium, or internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0044] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0045] As shown in Figure 1 and Figure 2 The intelligent control method of the data one-way transmission system based on LSTM prediction analysis according to the embodiments of the present application comprises the following steps:

[0046] S1: Collecting time series data affecting data transmission rate and data transmission success rate, and pre-processing the time series data to obtain a training set.

[0047] The time series data includes network I / O performance data, hardware I / O performance data and packet size of the sending end. In the embodiments of the present application, the network I / O performance data specifically includes network throughput, network bandwidth, propagation delay, queuing delay, processing delay, packet loss rate, network jitter, network congestion and network utilization. The hardware I / O performance data includes system load, process quantity, context switching, read-write delay, I / O throughput, computing throughput, concurrent throughput and memory throughput. The specific parameters of the time series data, the description corresponding to the parameters, the acquisition method corresponding to the parameters and the value range corresponding to the parameters are shown in Table 1:

[0048] Table 1:

[0049]

[0050] In the multi-performance type bidirectional interconnection system, data transmission is performed by using TCP (Transmission Control Protocol), and the average delay time of packet sending in the stable I / O performance range is obtained.

[0051] The preprocessing in step S1 comprises the following steps:

[0052] S11: The time series data is normalized by using the following formula to adapt to the input requirements of subsequent processing:

[0053] ;

[0054] wherein, denotes normalized data, denotes original data, and denotes original data corresponding maximum and minimum values.

[0055] For a given multivariate unidirectional transmission device performance time series data collected from different sources, the method proposed by the present application aims to predict the data transmission success rate under a specified parameter using a multivariate input feature sequence, so the collected time series data needs to be processed as follows:

[0056] S12: Organize the normalized time series data into a sequence according to the time , and further create a feature sequence is:

[0057] ;

[0058] corresponding data packet average delay time , denotes the data packet average delay time at time t, t denotes the time, and T denotes the total time;

[0059] Each element in the feature sequence contains K features, that is:

[0060] ;

[0061] In the embodiment of the present application, each element contains 18 features, that is, K=18.

[0062] S13: In order to meet the dimension required by the subsequent constructed neural network architecture, the feature sequence is further divided according to the time step n to obtain the training feature sequence :

[0063] ;

[0064] ;

[0065] wherein, m denotes the index of the subsequence in the training feature sequence , each training sample contains data from time m to m+n-1, denotes the total number of subsequences in the training feature sequence , and each subsequence ;

[0066] S14: training feature sequence and the corresponding packet average delay time to constitute a training set, the training feature sequence Subsequence corresponding Packet average delay time at time .

[0067] S2: build an LSTM-based neural network architecture, input the training set obtained in step S1 into the neural network architecture, train the neural network architecture, and obtain a network model.

[0068] LSTM is a popular deep learning technique for time series prediction in RNN (Recurrent Neural Network). For example, LSTM is not only used for classification and regression problems in stock market prediction, but also used in rainfall runoff modeling, fMRI (functional magnetic resonance imaging) data analysis, anomaly detection, mobile traffic prediction, etc. Although the standard RNN is superior to traditional networks in retaining information, it is not good at learning long-term dependencies due to the gradient vanishing problem. LSTM uses memory cells to overcome the problem of gradient vanishing, which consists of an input layer, a hidden layer, a cell state, and an output layer. The key component of the LSTM architecture is the cell state throughout the chain, which only has linear interaction and keeps the information flow unchanged.

[0069] Specifically, the LSTM-based neural network architecture includes an input layer, a hidden layer, and an output layer; wherein the input of the current time hidden layer includes the corresponding subsequence , the output of the previous time hidden layer and the state information of the previous time hidden layer; the output of the current time hidden layer includes the output of the current time hidden layer and the state information of the current time hidden layer; the output result of the last time hidden layer is output through the output layer after the full connection layer. The predicted packet average delay time.

[0070] The structure of the hidden layer is as shown in Figure 3 , including an input gate, a forget gate, an output gate, and a change gate. The gating mechanism of LSTM is used to delete or modify the cell state information, which is a method of selectively transmitting information. Among them:

[0071] The functions of the input gate include controlling the information flow, identifying the updated information, and combining the outputs of the change gate. The main task of the input gate in LSTM is to selectively control the update of information, ensuring that the network can effectively learn and remember long and short-term dependencies. By identifying which information needs to be updated, the input gate helps the model perform well in time series tasks. The implementation of the input gate is as follows:

[0072] ;

[0073] wherein represents the output of the input gate at time t, represents the weight matrix of the input gate, represents the bias matrix of the input gate, represents the weight matrix of the input gate acting on the output of the previous time hidden layer, represents the sigmoid activation operation;

[0074] The forget gate determines the degree of forgetting of the information of the previous time hidden layer by the current time hidden layer, i.e., obtaining information from the input subsequence and the output of the previous time hidden layer , and then producing an output between 0 and 1 through a sigmoid activation function, and then determining which information to discard from the state information of the previous time hidden layer : when the output value of the forget gate is 1, all information is stored in the current time hidden layer, and when the output value of the forget gate is 0, all information from the previous time hidden layer is forgotten, which can be represented as:

[0075] ;

[0076] wherein represents the output of the forget gate at time t, represents the weight matrix of the forget gate, represents the bias matrix of the forget gate, represents the weight matrix of the forget gate acting on the output of the previous time hidden layer;

[0077] The change gate determines the state information of the current time hidden layer according to the information of the forget gate and the input gate, i.e.:

[0078] ;

[0079] wherein represents the output of the change gate at time t, represents the weight matrix of the change gate, represents the bias matrix of the change gate, represents the weight matrix of the change gate acting on the output of the previous time hidden layer​​ a weight matrix on the hidden layer at the previous time step, denotes the hyperbolic tangent activation operation; the state information of the hidden layer at the current time is obtained by

[0080]

[0081] denotes element-wise multiplication of matrices, denotes the state information of the hidden layer at the previous time step;

[0082] The function of the output gate includes controlling the output information and combining the cell state. The main task of the output gate in the LSTM is to selectively output information according to the current cell state to generate the hidden state at the current time. The output gate helps the LSTM maintain and pass important time series data features by controlling which information is output, ensuring that the model can effectively handle long and short term dependencies, so that the LSTM performs well in many sequence modeling tasks. The implementation of the output gate is as follows:

[0083]

[0084] denotes the output of the output gate at time t, denotes a weight matrix of the output gate, denotes a bias matrix of the output gate, denotes a weight matrix of the output gate acting on the output of the hidden layer at the previous time step, denotes the output of the hidden layer at the current time,

[0085]

[0086] The data processing process of the hidden layer at the current time can also be represented by the following formula:

[0087]

[0088] denotes the data processing process of the hidden layer at the current time, denotes all parameters that need to be trained and learned by the hidden layer at the current time.

[0089] In the embodiment of the application, the optimizer in the training process of the neural network architecture is Adam. The Adam optimizer dynamically adjusts the learning rate during the training process, and has good convergence speed and effect. The default parameters of the Adam optimizer include: learning rate lr=0.001, exponential decay rate of first moment estimation of momentum , exponential decay rate of second moment estimation of momentum (exponential decay rate​​​​​​ and The weight for adjusting momentum, a numerical stability term ϵ=1e-7 (for preventing division by zero error). The time step in the training process of the neural network architecture is 10; the batch size is 64; the training rounds (Epochs) are 100. The loss function in the training process is the mean absolute error that is:

[0090] ;

[0091] wherein, represents the i-th predicted value, represents the i-th actual value, represents the total number of samples or records.

[0092] S3: input the time series data to be tested into the network model obtained in step S2 for prediction, and adjust the neural network architecture according to the prediction result; repeat step S2 to train the adjusted neural network architecture until the training is completed to obtain the best model.

[0093] In step S3, the time series data to be tested is input into the network model for prediction to obtain the predicted packet average delay time. In combination with the real packet average delay time corresponding to the time series data to be tested, the predicted packet average delay time is evaluated, and the hyperparameters (including learning rate, batch size, number of hidden layer units, etc.) of the neural network architecture are adjusted according to the evaluation result. In the embodiment of the present application, the evaluation indexes for evaluating the predicted packet average delay time include root mean square error (RMSE), correlation coefficient (R) and mean absolute error (MAE), wherein:

[0094] The root mean square error RMSE measures the error between the predicted value and the true value, and pays special attention to the larger error. It is the square root of the sum of squares of errors, so it emphasizes the importance of larger errors, and the specific formula is:

[0095] ;

[0096] The correlation coefficient R measures the linear relationship between the predicted value and the true value. R=1 indicates perfect positive correlation, and the model predicted value is completely consistent with the true value; R=0 indicates no correlation, and there is no linear relationship between the model predicted value and the true value; R=-1 indicates perfect negative correlation.

[0097] The specific formula is:

[0098] ;

[0099] wherein, and respectively represent the mean value of the true value and the predicted value; the mean absolute error MAE measures the average value of the absolute error between the predicted value and the true value, which represents the average accuracy of the model prediction, and the specific formula is:

[0100] .

[0101] S4: deploying the optimal model obtained in step S3 to the actual data unidirectional transmission system, and the unidirectional transmission system predicts the transmission delay time of the data packet according to the optimal model, and adjusts and sends the data packet according to the predicted transmission delay time.

[0102] The process of step S4 is shown in Figure 4 The unidirectional transmission system includes a first thread thread1 and a second thread thread2. The first thread thread1 is responsible for data transmission, that is, packetizing data through UDP (User Datagram Protocol), adjusting and sending data according to the current transmission strategy. The second thread thread2 is responsible for performance monitoring, that is, periodically monitoring the I / O parameters P (p1, p2, …, p 18 ), such as network and hardware performance indicators (such as network throughput, delay, etc.); the second thread thread2 intelligently infers and predicts the transmission delay time (t ) of the data packet based on the I / O parameters P, and passes the predicted transmission delay time (t ) to the first thread thread1, and then adjusts the data transmission parameters (such as data packet size and sending delay) in real time. The sending end periodically predicts the average delay time, and adjusts the transmission parameters such as data packet size and delay time in real time, to ensure the efficient operation of the unidirectional transmission system.

[0103] It should be understood that various forms of the flow shown above can be used to reorder, add or delete steps. For example, the steps described in the present disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions of the present disclosure can be achieved, which is not limited herein.

[0104] The above specific embodiments do not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1.A method for intelligent control of a data one-way transmission system based on LSTM prediction analysis, characterized in that: The method comprises the following steps: S1: collecting time series data affecting data transmission rate and data transmission success rate, and preprocessing the time series data to obtain a training set; The preprocessing in step S1 comprises the following steps: S11: normalizing the time series data; S12: Organize the normalized time series data into sequences according to time and create the feature sequence X as: ； corresponding packet average delay time , denotes the packet average delay time at time t, t denotes the time, T denotes the total time each element in the feature sequence K features, i.e.: each element in the feature sequence ； S13: obtaining the feature sequence of the target object in the target frame based on the feature sequence of the target object in the first frame and the feature sequence of the target object in the second frame According to the time step n, the training feature sequence is further divided to obtain the training feature sequence : ； ； wherein m represents the training feature sequence subsequences of indexes, wherein m represents the training feature sequence subsequences total number of sub-sequences, each sub-sequence ; S14: the training feature sequence and the corresponding packet average delay time constituting the training set, the training feature sequence subsequence in the training feature sequence corresponding to the packet average delay time at time n+m ; S2: constructing a neural network architecture based on LSTM, inputting the training set obtained in step S1 into the neural network architecture, training the neural network architecture, and obtaining a network model; the neural network architecture comprises an input layer, a hidden layer, and an output layer; wherein the input of the current time hidden layer comprises a sub-sequence corresponding to the current time , the output of the previous time hidden layer, and the state information of the previous time hidden layer; the output of the current time hidden layer comprises the output of the current time hidden layer and the state information of the current time hidden layer; and the output result of the hidden layer at the last time is output through the output layer after a full connection layer to output a predicted average packet delay time; S3: inputting the time series data to be tested into the network model obtained in step S2 for prediction, adjusting the neural network architecture according to the prediction result; repeating step S2 to train the adjusted neural network architecture until the training is completed to obtain an optimal model; in step S3, the time series data to be tested is inputted into the network model for prediction to obtain predicted packet average delay time; the predicted packet average delay time is evaluated in combination with the real packet average delay time corresponding to the time series data to be tested, and the hyperparameters of the neural network architecture are adjusted according to the evaluation result; S4: deploying the optimal model obtained in step S3 to an actual one-way data transmission system, the one-way transmission system predicting the transmission delay time of the data packet according to the optimal model, and adjusting and sending the data packet according to the predicted transmission delay time; In step S4, the one-way transmission system comprises a first thread and a second thread, the optimal model predicting the transmission delay time of the data packet in the second thread, and the predicted transmission delay time being passed to the first thread; the first thread packs data, adjusts and sends the data packet according to the predicted transmission delay time. 2.The intelligent control method of the data one-way transmission system based on LSTM prediction analysis according to claim 1, characterized in that: In step S1, the time series data comprises network I / O performance data, hardware I / O performance data and packet size of the sending end. 3.The intelligent control method of the data one-way transmission system based on LSTM prediction analysis according to claim 1, characterized in that: In step S2, the hidden layer comprises an input gate, a forgetting gate, an output gate and a change gate; the input gate determines the information of the current time hidden layer, the forgetting gate determines the forgetting degree of the information of the previous time hidden layer by the current time hidden layer; the change gate determines the state information of the current time hidden layer according to the information of the forgetting gate and the input gate; the output gate outputs the output of the current time hidden layer according to the state information of the current time hidden layer.

Images