A multi-layer encryption high-security data protection method

Abstract

The application belongs to the technical field of data protection, and discloses a high-safety data protection method with multi-layer encryption, which comprises the following steps: step 1, receiving encrypted remote sensing data sent by a UAV end; step 2, decrypting the encrypted remote sensing data and performing preprocessing; step 3, performing quantum encryption on the preprocessed remote sensing data, sending the remote sensing data to a storage end, and generating corresponding quantum keys; step 4, collecting historical influence coefficients of each sub-distribution path in a quantum key distribution path, and predicting the influence coefficients of each sub-distribution path at a future time based on the historical influence coefficients of each sub-distribution path; step 5, if any of the n-1 influence coefficients at the same future time exceeds a preset influence coefficient threshold, step 4 is executed until the n-1 influence coefficients at the same future time all do not exceed the preset influence coefficient threshold; and the integrity and confidentiality of the remote sensing data are effectively maintained at all times.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data protection, more particularly, the present application relates to a multi-layer encryption high security data protection method. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology, unmanned aerial vehicles are increasingly widely used in environmental monitoring, agricultural management, disaster assessment and other fields. These unmanned aerial vehicles are equipped with high-resolution remote sensing sensors and can collect a large amount of remote sensing data in real time. However, remote sensing data faces many security challenges during its collection, transmission, processing and storage on the unmanned aerial vehicle side; remote sensing data often involves sensitive information, and if it is illegally stolen or tampered with during transmission, it will cause serious security risks.

[0003] Although the prior art has made certain progress in ensuring the security of remote sensing data, there are still the following technical problems:

[0004] The instability of the quantum key distribution path causes the quantum key to be sent multiple times, the encryption system is forced to remain in a degraded mode, the original encryption expectations are reduced, resulting in no difference from ordinary encryption, and the transmission of remote sensing data cannot be guaranteed.

[0005] In view of this, the present application proposes a multi-layer encryption high security data protection method to solve the above problems. SUMMARY

[0006] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purpose, the present application provides the following technical scheme: a multi-layer encryption high security data protection method, the method comprising:

[0007] Step 1, receiving encrypted remote sensing data sent by the unmanned aerial vehicle side;

[0008] Step 2, decrypting the encrypted remote sensing data and performing preprocessing;

[0009] Step 3, quantum encrypting the preprocessed remote sensing data, sending it to the storage side, and generating the corresponding quantum key;

[0010] Step 4, collecting the historical influence coefficients of each sub-distribution path in the quantum key distribution path, and predicting the influence coefficients of each sub-distribution path at future time based on the historical influence coefficients of each sub-distribution path;

[0011] Step 5, if any of the n-1 influence coefficients at the same future time exceeds the preset influence coefficient threshold, step 4 is executed, until the n-1 influence coefficients at the same future time do not exceed the preset influence coefficient threshold, and step 6 is executed;

[0012] Step 6: Input the influence coefficient of each sub-distribution path at the future moment into the time prediction model one by one to obtain the required distribution time corresponding to each sub-distribution path at the future moment, where the required distribution time is the time required for the quantum key to pass through the sub-distribution path;

[0013] Step 7: If the sum of the n-1 required distribution times does not exceed the preset required distribution time threshold, calculate the quantum key distribution time according to the future time and the sum of the n-1 required distribution times, and distribute the quantum key to the storage node via the quantum key distribution path when the quantum key distribution time is reached.

[0014] Furthermore, the method of predicting the influence coefficient of each sub-distribution path at a future moment based on the historical influence coefficient of each sub-distribution path includes:

[0015] Input the historical impact coefficient of each sub-distribution path into the pre-trained prediction model to obtain the impact coefficient of each sub-distribution path at the future moment;

[0016] The training method of the prediction model includes:

[0017] Collect the historical impact coefficients of each sub-distribution path. The historical impact coefficients include m time-series-based impact coefficients, where m is an integer greater than 1. Preset the prediction time step T, sliding step P2, and sliding window length P1. Use the sliding window method to convert the m time-series-based impact coefficients into multiple training samples, with each training sample corresponding to one label, to form a set of training data.

[0018] The training data is used as the input of the prediction model, the impact coefficient of the future moment after the predicted time step T is used as the output, the actual impact coefficient of the future moment after each time step T is used as the prediction target, and the preset accuracy is used as the training target. The prediction model is trained to obtain a prediction model that meets the preset accuracy. The prediction model is a long short-term memory network or a support vector machine.

[0019] Furthermore, the influence coefficient generation method includes:

[0020] Collect the influencing parameters of n-1 sub-distribution paths of n communication nodes on the quantum key distribution path;

[0021] Based on the influence parameters of the n-1 sub-distribution paths, generate the influence coefficients of the one-to-one corresponding sub-distribution paths;

[0022] The n communication nodes consist of edge computing nodes, communication relay nodes, and storage nodes. The n-1 sub-distribution paths include the optical sub-distribution path between the edge computing node and its adjacent communication relay node, the optical fiber sub-distribution path between the communication relay node and its adjacent communication relay node, and the optical fiber sub-distribution path between the storage node and its adjacent communication relay node.

[0023] The influence coefficient includes a first influence coefficient and n-2 second influence coefficients; the first influence coefficient is the influence coefficient of the optical path sub-distribution path; the second influence coefficient is the influence coefficient of the optical fiber sub-distribution path.

[0024] Furthermore, the parameters affecting the optical distribution path include photon loss rate PL, quantum bit error rate QBER, environmental noise level EN and transmission distance TD;

[0025] The ambient noise level EN is calculated as follows:

[0026] EN=α temp ×T+α humidity ×H+α wind ×W;

[0027] Where T is temperature, H is humidity, W is wind speed, and α temp , α humidity , α wind is the corresponding weight.

[0028] Furthermore, the method for generating the first influencing parameter C1 includes:

[0029]

[0030] Where w1, w2, w3, w4, w5, w6, w7, and w8 are the corresponding weights.

[0031] Furthermore, the influencing parameters of the optical fiber sub-distribution path include the optical fiber loss rate FQ, the optical fiber length FLT, the communication relay node performance score RNP and the interference intensity MVI;

[0032] The communication relay node performance score RNP is calculated as follows:

[0033]

[0034] Where RN is the average quantum bit error rate per unit time of the communication relay node; RP is the average processing delay per unit time of the communication relay node;

[0035] The interference intensity MVI is calculated as follows:

[0036] MVI=δ vib ×V+δ emi ×I;

[0037] Where V is the vibration intensity, I is the electromagnetic interference intensity, δ vib , δ emi is the corresponding weight.

[0038] Furthermore, the method for generating the second influencing parameter C2 includes:

[0039]

[0040] Where v1, v2, v3, v4, v5, v6, v7, v8, and v9 are the corresponding weights.

[0041] Furthermore, the training method of the time prediction model includes:

[0042] Pre-collecting the required distribution time training data of group A, where A is an integer greater than 1, and the required distribution time training data includes the influence coefficient and the required distribution time corresponding to the influence coefficient;

[0043] Each group of influence coefficients is used as the input of a time prediction model. The time prediction model uses the required distribution time corresponding to each group of influence coefficients as the output, the actual required distribution time corresponding to each group of influence coefficients as the prediction target, and minimizing the sum of the loss function values ​​of all influence coefficients as the training target; the time prediction model is trained until the sum of the prediction errors reaches convergence and the training is stopped; the time prediction model is specifically a deep neural network model.

[0044] Furthermore, the loss function value of the time prediction model is as follows:

[0045]

[0046] Where J(θ) is the loss function value; q θ (x(a)) is the prediction function of the time prediction model, based on the parameter θ on the a-th influence coefficient x (a) Make prediction output; y (a) is the actual output corresponding to the ath influence coefficient, that is, the required distribution time, x (a) is the ath influence coefficient; θ is the parameter of the time prediction model, namely the influence coefficient weight and bias, which is optimized by stochastic gradient descent algorithm training;

[0047] Methods for optimizing the parameters θ of the time prediction model through stochastic gradient descent training include:

[0048] Each time, an influence coefficient is randomly extracted from the required distribution time training data, and the parameters are updated according to the gradient of the loss function with respect to the influence coefficient;

[0049] At the tth update:

[0050]

[0051] θ t+1 =θ t -v t ;

[0052] Where t is the number of iterations of the time prediction model; π is the momentum coefficient, which is modified based on the complexity of the model; v t is the momentum variable of the tth iteration; v t-1 The momentum variable of the t-1th iteration; v1 is the momentum variable of the 1st iteration; is the gradient of the loss function value of the time prediction model with respect to the parameter θ; t+1 is the parameter of the loss function value of the time prediction model at the t+1th iteration; θ t The parameters of the loss function value of the time prediction model at the tth iteration, including the weight and bias of the time prediction model.

[0053] Furthermore, methods for modifying the momentum coefficient π based on model complexity include:

[0054] Step 1: Evaluate the model complexity, traverse each layer of the time prediction model, and obtain parameter data N θ , the number of layers N that the time prediction model passes through from the input layer to the output layer c and the data N of the nonlinear operation f ;

[0055] Calculate the model complexity Γ:

[0056] Γ=w θ N θ +w c N c +w f N f ;

[0057] Among them, w θ , w c and w f N θ , N c and N f The corresponding weight coefficient;

[0058] Step 2: Correct the momentum coefficient and calculate the attenuation factor ψ(Γ) of the momentum coefficient:

[0059]

[0060] Among them, e is a mathematical constant; ξ is the decay rate control parameter; Γ th is the preset model complexity threshold; calculate the momentum coefficient: π=π0×ψ(Γ); where π0 is the preset original momentum coefficient;

[0061] Repeat the stochastic gradient descent algorithm to train the parameters θ of the optimization time prediction model until the loss function converges; the required distribution time output when the loss function converges is used as the final output result.

[0062] The beneficial effects of the multi-layer encryption high-security data protection method of the present invention are as follows:

[0063] After preliminary encryption, remote sensing data collected by the drone is first transmitted to the edge computing end. The edge computing end is responsible for decrypting and preprocessing this remote sensing data, then implementing quantum encryption, sending the encrypted data to the storage node, and generating the corresponding quantum key. Before distributing the quantum key, the system collects the historical influence coefficients of each sub-distribution path in the quantum key distribution path and uses this historical data to predict the influence coefficient of each sub-distribution path at a future time.

[0064] If the influence coefficient of any of the n-1 sub-distribution paths at the same future moment exceeds the preset threshold (this indicates that a communication relay node may be attacked, which may cause quantum key distribution to be blocked and fail with a high probability), the system will re-execute the preliminary analysis module until the influence coefficients of all n-1 sub-distribution paths at that future moment do not exceed the preset threshold, ensuring a high success rate of quantum key distribution.

[0065] The system then inputs the influence coefficient of each sub-distribution path at a future moment into a time prediction model to calculate the required distribution time for each path at that future moment. These required distribution times represent the time it takes for the quantum key to traverse each sub-distribution path. If the total required distribution time for the n-1 distribution paths does not exceed a preset time threshold, the system determines the quantum key distribution moment based on the future moment and the total required distribution time for the n-1 distribution paths, and securely distributes the quantum key to the storage node at that moment.

[0066] Through this multi-layer encryption mechanism, the system effectively improves the success rate and security of quantum key distribution, avoids encryption degradation caused by the instability of the quantum distribution path, and comprehensively guarantees the security of remote sensing data during each transmission process. In this embodiment, the remote sensing data undergoes multiple encryption processes during transmission, forming a rigorous multi-layer protection system. This ensures that the data remains highly secure during transmission between the drone end, the edge computing end, and the storage node, preventing potential network attacks and data leakage risks.

[0067] The advantages of high-security data protection with multi-layer encryption are as follows:

[0068] Initial encryption: Remote sensing data is encrypted for the first time on the drone side, ensuring basic confidentiality before being transmitted to the edge computing side. Quantum encryption: The edge computing side implements quantum encryption after preprocessing, utilizing quantum key distribution (QKD) technology to provide theoretically unbreakable encryption strength, significantly improving data security. Dynamic key management: By predicting and monitoring the influence coefficient of the distribution path, it dynamically adjusts the quantum key distribution strategy to avoid potential attacks and improve the reliability of key distribution. Multiple encrypted transmissions: Remote sensing data is encrypted multiple times during transmission between different nodes, forming a multi-layered protection barrier. Even if one layer is breached, the other layers can still ensure data security. Stability and efficiency: Combining time prediction models to optimize the required distribution time ensures that quantum keys are distributed at the optimal time, improving both security and the efficiency and stability of the distribution process.

[0069] Through multi-layer encryption and dynamic management strategies, the system not only provides solid security protection in all aspects of data transmission, but also demonstrates excellent protection capabilities and adaptability when facing complex and changing network environments, ensuring that the integrity and confidentiality of remote sensing data are always effectively maintained. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a schematic diagram of the connection between the drone end, edge computing end, communication relay node, and storage node in the present invention;

[0071] Figure 2 A schematic diagram of a multi-layer encryption high-security data protection system according to the present invention;

[0072] Figure 3 This is a flow chart of a multi-layer encryption high-security data protection method of the present invention. DETAILED DESCRIPTION

[0073] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0074] Example 1

[0075] See Figure 1-2 The multi-layer encryption high-security data protection system described in this embodiment is applied in an edge computing terminal; the edge computing terminal includes:

[0076] The receiving module is used to receive encrypted remote sensing data sent by the drone. Remote sensing data is collected by various sensors installed on the drone. It should be noted that other network data, besides remote sensing data, can also be used. Encrypted remote sensing data is encrypted by the drone using the AES algorithm and a one-time key is generated by combining the drone's device ID and timestamp to ensure secure data transmission between the drone and the edge computing device, which is a computer device located near the drone.

[0077] The data processing module is used to decrypt the encrypted remote sensing data and perform preprocessing to reduce the computing pressure on the drone side. The preprocessing includes data cleaning, data integration, data reduction and other operations.

[0078] The quantum encryption module is used to perform quantum encryption on the pre-processed remote sensing data, send it to the storage end, and generate the corresponding quantum key.

[0079] The first analysis module collects the historical influence coefficient of each sub-distribution path in the quantum key distribution path, and predicts the influence coefficient of each sub-distribution path at a future moment based on the historical influence coefficient of each sub-distribution path.

[0080] The method for predicting the influence coefficient of each sub-distribution path at a future moment based on the historical influence coefficient of each sub-distribution path includes:

[0081] The historical impact coefficient of each sub-distribution path is input into the pre-trained prediction model to obtain the impact coefficient of each sub-distribution path at the future moment.

[0082] The training methods for the prediction model include:

[0083] Collect the historical impact coefficients of each sub-distribution path. The historical impact coefficients include m impact coefficients arranged based on time series, where m is an integer greater than 1. Preset the prediction time step T, sliding step P2, and sliding window length P1. Use the sliding window method to convert the m impact coefficients arranged based on time series into multiple training samples. One training sample corresponds to one label, and constitute a set of training data.

[0084] The training data is used as the input of the prediction model, the impact coefficient of the future moment after the predicted time step T is used as the output, the actual impact coefficient of the future moment after each time step T is used as the prediction target, and the preset accuracy is used as the training target. The prediction model is trained to obtain a prediction model that meets the preset accuracy. The prediction model is a long short-term memory network or a support vector machine.

[0085] The methods for generating influence coefficients include:

[0086] Collect the influencing parameters of n-1 sub-distribution paths of n communication nodes on the quantum key distribution path;

[0087] Based on the influence parameters of the n-1 sub-distribution paths, generate the influence coefficients of the one-to-one corresponding sub-distribution paths;

[0088] It should be noted that the n communication nodes are composed of edge computing nodes, communication relay nodes, and storage nodes, and the n-1 sub-distribution paths include the optical path sub-distribution path between the edge computing node and its adjacent communication relay node, the optical fiber sub-distribution path between the communication relay node and its adjacent communication relay node, and the optical fiber sub-distribution path between the storage node and its adjacent communication relay node.

[0089] The influence coefficient includes a first influence coefficient and n-2 second influence coefficients; the first influence coefficient is the influence coefficient of the optical path sub-distribution path; the second influence coefficient is the influence coefficient of the optical fiber sub-distribution path.

[0090] The parameters affecting the optical distribution path include photon loss rate PL, quantum bit error rate QBER, environmental noise level EN and transmission distance TD.

[0091] The photon loss rate (PL) is the ratio of photons lost during transmission. An optical power meter is used to monitor the optical power at the transmitter and receiver in real time to calculate the loss rate. Photon loss directly affects the key distribution rate and security. The photon loss rate (PL) is calculated as follows:

[0092]

[0093] In the formula, optical power meters are installed at the transmitting and receiving ends to monitor the transmitted optical power P in real time. send and the received optical power P recv ; Calculate the photon loss rate.

[0094] The qubit error rate (QBER) is the ratio of bit errors during transmission. It is calculated by comparing the sent and received bit sequences. QBER is an important indicator for evaluating the security of QKD. Excessively high QBER may indicate potential eavesdropping. The qubit error rate (QBER) is calculated as follows:

[0095]

[0096] In the formula, the number of error bits E and the total number of bits N are counted by comparing the bit sequences at the sending and receiving ends.

[0097] The environmental noise level (EN) affects the intensity of environmental noise in the quantum signal, such as temperature, humidity, and wind speed. This noise is monitored and acquired using environmental sensors installed near the optical path. The environmental noise level will increase the QBER and affect the security and efficiency of QKD. The environmental noise level (EN) is calculated as follows:

[0098] EN=αtemp ×T+α humidity ×H+α wind ×W;

[0099] Where T is temperature, H is humidity, W is wind speed, and α temp , α humidity , α wind is the corresponding weight.

[0100] The transmission distance TD is the transmission distance of the optical distribution path between the edge computing node and the adjacent communication relay node. The distance is calculated by measuring the starting and ending points of the optical distribution path. The transmission distance affects the photon loss and QBER. The longer the distance, the greater the loss.

[0101] The method for generating the first influencing parameter C1 includes:

[0102]

[0103] Where, PL × QBER: reflects the interaction between photon loss rate and bit error rate. Higher loss will lead to increased bit error rate. EN 2 : The nonlinear effect of environmental noise on the security of quantum key distribution (QKD) is taken into account. Higher noise will significantly increase the QBER; log(TD+1): The transmission distance is logarithmically transformed to reduce the excessive amplification of the coefficients due to long-distance transmission; 1+w7×TD+w8×(TD) 2 : Normalization processing is used to prevent the transmission distance from being amplified unlimitedly by C1; w1, w2, w3, w4, w5, w6, w7, and w8 are the corresponding weights.

[0104] The influencing parameters of the optical fiber sub-distribution path include the optical fiber loss rate FQ, the optical fiber length FLT, the communication relay node performance score RNP and the interference intensity MVI.

[0105] Fiber loss (FQ) is the optical loss per kilometer of fiber, expressed in dB / km. Fiber loss characteristics are measured using an optical time domain reflectometer (OTDR). Fiber loss affects transmission loss and the quantum bit error rate (QBER). Fiber length affects both the quantum bit error rate (QBER) and transmission efficiency.

[0106] The communication relay node performance score (RNP) reflects the device status of the communication relay node, including the bit error rate and delay time of the monitoring optical module. The performance of the communication relay node directly affects the signal transmission quality and quantum bit error rate (QBER). The communication relay node performance score is calculated as follows:

[0107]

[0108] Where RN is the average quantum bit error rate per unit time of the communication relay node; RP is the average processing delay per unit time of the communication relay node.

[0109] The interference intensity MVI is the mechanical vibration and electromagnetic interference intensity in the optical fiber transmission path; vibration and electromagnetic sensors are used to monitor the vibration and electromagnetic interference intensity near the optical fiber; the mechanical vibration and electromagnetic interference intensity will cause signal distortion and increase the quantum bit error rate QBBR.

[0110] MVI=δ vib ×V+δ emi ×I;

[0111] Where V is the vibration intensity, I is the electromagnetic interference intensity, δ vib , δ emi is the corresponding weight.

[0112] The second influencing parameter C2 generation method includes:

[0113]

[0114] Where, FL×FLT: reflects the interaction between optical fiber loss rate and optical fiber length. Longer optical fiber will increase optical loss. RNP 2 : Capturing the nonlinear impact of communication relay node performance, high-performance communication relay nodes can significantly improve system security; Alleviate the impact of high-intensity interference to make the formula more adaptable; 1+v8×FLT+v9×sin(FLT): normalization processing to prevent the fiber length from being increased without limit; v1, v2, v3, v4, v5, v6, v7, v8, and v9 are corresponding weights.

[0115] In the second analysis module, if any of the n-1 influence coefficients at the same future moment exceeds the preset influence coefficient threshold (indicating that a certain communication relay node may be attacked, and at this time the quantum key distribution will probably be affected and will most likely fail), the first analysis module will be repeatedly executed until the n-1 influence coefficients at the same future moment do not exceed the preset influence coefficient threshold (at this time the quantum key distribution has a high success rate), and the third analysis module will be executed.

[0116] The third analysis module inputs the influence coefficient of each sub-distribution path at the future moment into the time prediction model one by one to obtain the required distribution time corresponding to each sub-distribution path at the future moment. The required distribution time is the time required for the quantum key to pass through the sub-distribution path.

[0117] The training methods for time prediction models include:

[0118] Pre-collecting A-group required distribution time training data, A is an integer greater than 1, and the required distribution time training data includes an impact coefficient and a required distribution time corresponding to the impact coefficient.

[0119] Taking each group of impact coefficients as an input of a time prediction model, taking a required distribution time corresponding to each group of impact coefficients as an output of the time prediction model, taking an actual required distribution time corresponding to each group of impact coefficients as a prediction target, and taking a sum of loss function values of all impact coefficients as a training target, the time prediction model is trained until a sum of prediction errors reaches convergence to stop training; and the time prediction model is specifically a deep neural network model.

[0120] The loss function value of the time prediction model is as follows:

[0121]

[0122] In the formula, J(θ) is a loss function value; A is a number of required distribution time training data; q θ (x (a) ) is a prediction function of the time prediction model, and is used for predicting an output based on a parameter θ and an a-th impact coefficient x (a) ; y (a) is an actual output corresponding to the a-th impact coefficient, that is, a required distribution time, x (a) is the a-th impact coefficient; and θ is a parameter of the time prediction model, that is, an impact coefficient weight and a bias, which is optimized by a stochastic gradient descent algorithm.

[0123] The method for optimizing the parameter θ of the time prediction model by the stochastic gradient descent algorithm comprises the following steps:

[0124] Randomly extracting an impact coefficient from the required distribution time training data each time, and updating the parameter according to a gradient of the loss function with respect to the impact coefficient.

[0125] At the t-th update:

[0126]

[0127] θ t+1 = θ t -v t ;

[0128] Wherein, t is an iteration number of the time prediction model; π is a momentum coefficient, and the momentum coefficient π is corrected based on model complexity; v t is a momentum variable of the t-th iteration; v t-1 is a momentum variable of the t-1th iteration; and v1 is a momentum variable of the 1st iteration. is a gradient of the loss function value of the time prediction model with respect to the parameter θ; and θt+1 is the parameter of the loss function value of the time prediction model at the t+1th iteration; θ t The parameters of the loss function value of the time prediction model at the tth iteration, including the weight and bias of the time prediction model.

[0129] Methods for modifying the momentum coefficient π based on model complexity include:

[0130] Step 1: Evaluate the model complexity, traverse each layer of the time prediction model, and obtain parameter data N θ , the number of layers N that the time prediction model passes through from the input layer to the output layer c and the data N of the nonlinear operation f .

[0131] Calculate the model complexity Γ:

[0132] Γ=w θ N θ +w c N c +w f N f

[0133] Among them, w θ , w c and w f N θ , N c and N f The corresponding weight coefficient.

[0134] Step 2: Correct the momentum coefficient and calculate the attenuation factor ψ(Γ) of the momentum coefficient:

[0135]

[0136] Among them, e is a mathematical constant; ξ is the decay rate control parameter; Γ th is the preset model complexity threshold; calculate the momentum coefficient: π=π0×ψ(Γ); where π0 is the preset original momentum coefficient.

[0137] Repeat the stochastic gradient descent algorithm to train the parameters θ of the optimization time prediction model until the loss function converges; the required distribution time output when the loss function converges is used as the final output result.

[0138] Using the stochastic gradient descent (SGD) algorithm to optimize the parameters θ of the time prediction model not only accelerates model training but also efficiently processes large amounts of data while avoiding local optima to a certain extent. During the training of the time prediction model, the SGD algorithm, combined with a momentum method that considers model complexity, dynamically adjusts the momentum coefficient based on model complexity. When model complexity is high, lowering the momentum coefficient reduces the cumulative impact of historical gradients, making parameter updates more cautious. This helps prevent excessive parameter updates caused by excessive momentum, making the model training process more stable and enabling better convergence when processing complex data. Furthermore, the model-complexity-aware momentum method optimizes the entire training process by dynamically adjusting the momentum. As model complexity increases, this avoids the accumulation of invalid historical gradients, allowing the model to adapt more quickly to new gradient changes. This improves training efficiency, shortens training time, and more quickly achieves ideal time prediction model performance.

[0139] In the fourth analysis module, if the sum of the n-1 required distribution times does not exceed the preset required distribution time threshold, the quantum key distribution time is calculated based on the future time and the sum of the n-1 required distribution times. When the quantum key distribution time is reached, the quantum key is distributed to the storage node via the quantum key distribution path; the storage node decrypts the remote sensing data according to the quantum key, and then distributes the data to the storage node and encrypts it, thereby improving the disaster recovery and anti-attack capabilities of the remote sensing data.

[0140] The remote sensing data collected by the unmanned aerial vehicle end is transmitted to the edge computing end after initial encryption, the remote sensing data is decrypted and preprocessed by the edge computing end, then quantum encryption is performed, and the remote sensing data is sent to the storage node, and the corresponding quantum key is generated; before distributing the quantum key, the historical influence coefficients of each sub-distribution path in the quantum key distribution path are collected, and the influence coefficients of each sub-distribution path at a future time are predicted based on the historical influence coefficients of each sub-distribution path; if any one of the n-1 influence coefficients at the same future time exceeds the preset influence coefficient threshold (indicating that a certain communication relay node may be attacked, at this time the quantum key distribution may be affected and may fail with a high probability), the first analysis module is repeatedly executed until the n-1 influence coefficients at the same future time do not exceed the preset influence coefficient threshold (at this time the success rate of quantum key distribution is high); the influence coefficients of each sub-distribution path at a future time are input into a time prediction model one by one, and the required distribution time corresponding to each sub-distribution path at a future time is obtained, the required distribution time being the time required for the quantum key to pass through the sub-distribution path; if the sum of the n-1 required distribution times does not exceed the preset required distribution time threshold, the quantum key distribution time is calculated according to the future time and the sum of the n-1 required distribution times, and the quantum key is distributed to the storage node when the quantum key distribution time is reached, thereby effectively improving the success rate and security of distributing the quantum key, avoiding the instability of the quantum distribution path and causing encryption degradation, and ensuring the security of the remote sensing data in each transmission process.

[0141] Embodiment 2

[0142] Referring to Figure 3 The embodiment provides a multi-layer encryption high-security data protection method, which is applied to an edge computing end, and the method comprises the following steps:

[0143] Step 1, receiving encrypted remote sensing data sent by the unmanned aerial vehicle end;

[0144] Step 2, decrypting and preprocessing the encrypted remote sensing data;

[0145] Step 3, quantum encrypting the preprocessed remote sensing data, sending the remote sensing data to a storage end, and generating a corresponding quantum key;

[0146] Step 4, collecting the historical influence coefficients of each sub-distribution path in the quantum key distribution path, and predicting the influence coefficients of each sub-distribution path at a future time based on the historical influence coefficients of each sub-distribution path;

[0147] Step 5, if any one of the n-1 influence coefficients at the same future time exceeds the preset influence coefficient threshold, step 4 is executed until the n-1 influence coefficients at the same future time do not exceed the preset influence coefficient threshold, and step 6 is executed;

[0148] Step 6: Input the influence coefficient of each sub-distribution path at the future moment into the time prediction model one by one to obtain the required distribution time corresponding to each sub-distribution path at the future moment, where the required distribution time is the time required for the quantum key to pass through the sub-distribution path;

[0149] Step 7: If the sum of the n-1 required distribution times does not exceed the preset required distribution time threshold, calculate the quantum key distribution time according to the future time and the sum of the n-1 required distribution times, and distribute the quantum key to the storage node via the quantum key distribution path when the quantum key distribution time is reached.

[0150] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

[0151] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A high-security data protection method with multi-layer encryption, characterized in that: Applied in edge computing nodes, the methods include: Step 1: Receive encrypted remote sensing data sent by the drone; Step 2: decrypt the encrypted remote sensing data and perform preprocessing; Step 3: Perform quantum encryption on the pre-processed remote sensing data, send it to the storage node, and generate the corresponding quantum key; Step 4: Collect the historical influence coefficient of each sub-distribution path in the quantum key distribution path, and predict the influence coefficient of each sub-distribution path at a future moment based on the historical influence coefficient of each sub-distribution path; Step 5: If any of the n-1 influence coefficients at the same future time exceeds the preset influence coefficient threshold, execute step 4. This process continues until none of the n-1 influence coefficients at the same future time exceeds the preset influence coefficient threshold, and then execute step 6. Step 6: Input the influence coefficient of each sub-distribution path at the future moment into the time prediction model one by one to obtain the required distribution time corresponding to each sub-distribution path at the future moment, where the required distribution time is the time required for the quantum key to pass through the sub-distribution path; Step 7: If the sum of the n-1 required distribution times does not exceed the preset required distribution time threshold, calculate the quantum key distribution time according to the future time and the sum of the n-1 required distribution times, and distribute the quantum key to the storage node via the quantum key distribution path when the quantum key distribution time is reached.

2. A high-security data protection method with multi-layer encryption according to claim 1, characterized in that: The method for predicting the influence coefficient of each sub-distribution path at a future moment based on the historical influence coefficient of each sub-distribution path includes: Input the historical impact coefficient of each sub-distribution path into the pre-trained prediction model to obtain the impact coefficient of each sub-distribution path at the future moment; The training method of the prediction model includes: Collect the historical impact coefficients of each sub-distribution path. The historical impact coefficients include m time-series-based impact coefficients, where m is an integer greater than 1. Preset the prediction time step T, sliding step P2, and sliding window length P1. Use the sliding window method to convert the m time-series-based impact coefficients into multiple training samples, with each training sample corresponding to one label, to form a set of training data. The training data is used as the input of the prediction model, the impact coefficient of the future moment after the predicted time step T is used as the output, the actual impact coefficient of the future moment after each time step T is used as the prediction target, and the preset accuracy is used as the training target. The prediction model is trained to obtain a prediction model that meets the preset accuracy. The prediction model is a long short-term memory network or a support vector machine.

3. A high-security data protection method with multi-layer encryption according to claim 2, characterized in that: The influence coefficient generation method includes: Collect the influencing parameters of n-1 sub-distribution paths of n communication nodes on the quantum key distribution path; Based on the influence parameters of the n-1 sub-distribution paths, generate the influence coefficients of the one-to-one corresponding sub-distribution paths; The n communication nodes consist of edge computing nodes, communication relay nodes, and storage nodes. The n-1 sub-distribution paths include the optical sub-distribution path between the edge computing node and its adjacent communication relay node, the optical fiber sub-distribution path between the communication relay node and its adjacent communication relay node, and the optical fiber sub-distribution path between the storage node and its adjacent communication relay node. The influence coefficient includes a first influence coefficient and n-2 second influence coefficients; the first influence coefficient is the influence coefficient of the optical path sub-distribution path; the second influence coefficient is the influence coefficient of the optical fiber sub-distribution path.

4. A multi-layer encryption high-security data protection method according to claim 3, characterized in that: The parameters affecting the optical distribution path include photon loss rate PL, quantum bit error rate QBER, environmental noise level EN and transmission distance TD; The ambient noise level EN is calculated as follows: EN=α temp ×T+α humidity ×H+α wind ×W; Where T is temperature, H is humidity, W is wind speed, and α temp is the temperature weight, α humidity is the humidity weight, α wind is the wind speed weight.

5. A multi-layer encryption high-security data protection method according to claim 4, characterized in that: The influencing parameters of the optical fiber sub-distribution path include the optical fiber loss rate FQ, optical fiber length FLT, communication relay node performance score RNP and interference intensity MVI; The communication relay node performance score RNP is calculated as follows: Where RN is the average quantum bit error rate per unit time of the communication relay node; RP is the average processing delay per unit time of the communication relay node; The interference intensity MVI is calculated as follows: MVI=δ vib ×V+δ emi ×I; Where V is the vibration intensity, I is the electromagnetic interference intensity, δ vib is the vibration intensity weight, δ emi is the electromagnetic interference intensity weight.

6. The high-security data protection method of multi-layer encryption according to claim 1, characterized in that: The training method of the time prediction model includes: Pre-collecting the required distribution time training data of group A, where A is an integer greater than 1, and the required distribution time training data includes the influence coefficient and the required distribution time corresponding to the influence coefficient; Each group of influence coefficients is used as the input of a time prediction model, and the time prediction model uses the required distribution time corresponding to each group of influence coefficients as the output, the required distribution time corresponding to each group of influence coefficients as the prediction target, and minimizing the sum of the loss function values ​​of all influence coefficients as the training target; the time prediction model is trained until the sum of the prediction errors reaches convergence and the training is stopped; the time prediction model is a deep neural network model.

7. A multi-layer encryption high-security data protection method according to claim 6, characterized in that: The loss function value of the time prediction model is as follows: Where J(θ) is the loss function value; q θ (x (a) ) is the prediction function of the time prediction model, based on the parameter θ on the a-th influence coefficient x (a) Make prediction output; y (a) is the output corresponding to the ath influence coefficient, that is, the required distribution time, x (a) is the ath influence coefficient; θ is the parameter of the time prediction model, namely the influence coefficient weight and bias, which is optimized by stochastic gradient descent algorithm training; Methods for optimizing the parameters θ of the time prediction model through stochastic gradient descent training include: Each time, an influence coefficient is randomly extracted from the required distribution time training data, and the parameters are updated according to the gradient of the loss function with respect to the influence coefficient; At the tth update: i t+1 =θ t -v t ; Where t is the number of iterations of the time prediction model; π is the momentum coefficient, which is modified based on the complexity of the model; v t is the momentum variable of the tth iteration; v t-1 The momentum variable of the t-1th iteration; v1 is the momentum variable of the 1st iteration; is the gradient of the loss function value of the time prediction model with respect to the parameter θ; t+1 is the parameter of the loss function value of the time prediction model at the t+1th iteration; θ t The parameters of the loss function value of the time prediction model at the tth iteration, including the weight and bias of the time prediction model.

8. A multi-layer encryption high-security data protection method according to claim 7, characterized in that: Methods for modifying the momentum coefficient π based on model complexity include: Step 1: Evaluate the model complexity, traverse each layer of the time prediction model, and obtain parameter data N θ , the number of layers N that the time prediction model passes through from the input layer to the output layer c and the data N of the nonlinear operation f ; Calculate the model complexity Γ: Γ=w θ N θ +w c N c +w f N f ; Among them, w θ , w c and w f N θ , N c and N f The corresponding weight coefficient; Step 2: Correct the momentum coefficient and calculate the attenuation factor ψ(Γ) of the momentum coefficient: Among them, e is a mathematical constant; ξ is the decay rate control parameter; Γ th is the preset model complexity threshold; calculate the momentum coefficient: π=π0×ψ(Γ); where π0 is the preset original momentum coefficient; Repeat the stochastic gradient descent algorithm to train the parameters θ of the optimization time prediction model until the loss function converges; the required distribution time output when the loss function converges is used as the final output result.

Images