Optimization method of offshore distributed power distribution unit unmanned aerial vehicle cooperative security communication network

Abstract

The invention discloses an optimization method of an offshore distributed power distribution unit unmanned aerial vehicle cooperative security communication network. The optimization method comprises the following steps: modeling a multi-communication link channel; modeling a signal transmission strategy; modeling a signal eavesdropping risk link, and evaluating the eavesdropping channel capacity of two scheduling links from a control center to a potential eavesdropper and from an unmanned aerial vehicle relay to the potential eavesdropper; formulating an information security problem; designing a scheduling problem optimization scheme; and outputting optimized power distribution network dispatching parameters, including a transmission power matrix, an unmanned aerial vehicle track and a power distribution unit grouping parameter. A power optimization algorithm and a track optimization algorithm are designed, the communication rate is improved, the safety and reliability of a power distribution unit dispatching system are maximized, and the operation management requirement of a power distribution network for safe access of offshore large-scale distributed energy sources is met.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communication networks of power systems, and in particular relates to an optimization method for an offshore distributed distribution unit unmanned aerial vehicle collaborative safety communication network. Background Art

[0002] With the large-scale development and utilization of renewable energy, offshore distributed energy access has become an important direction of energy transformation. However, due to the particularity of the offshore environment and the distributed characteristics of the distribution network, there are still many challenges in achieving safe and efficient access to offshore distributed energy. Due to the complex and changeable offshore environment, the construction cost of wired networks is high, the construction is difficult, and they are easily damaged in bad weather, resulting in insufficient communication reliability.

[0003] Although traditional wireless communication solutions are flexible, the existing communication mode based on fixed sites often cannot meet the dynamic scheduling requirements when facing large-scale distributed distribution units. The communication coverage is limited and the signal transmission stability is poor. In addition, in multi-user distributed systems, the existing communication protocols are inefficient in spectrum resource allocation and interference management, and it is difficult to meet the real-time and reliability requirements of dynamic control of large-scale offshore distribution networks. Especially in terms of security, since the offshore communication environment is exposed to open space and is vulnerable to eavesdropping and attacks, the existing technology often sacrifices part of the communication rate when protecting the communication link. This obviously cannot meet the demand for efficient and secure communication for offshore distributed energy access.

[0004] At the same time, the traditional UAV relay communication mode does not take into account the communication environment of the composite channel at sea. The UAV and power supply equipment are affected by the wind and waves at sea, and the complex channel needs to be modeled and optimized. Furthermore, due to the uncertainty of the marine environment, it is impossible to accurately know the specific location and channel state information of the eavesdropper, and it is necessary to optimize the communication power and UAV trajectory based on the fuzzy channel state information of the eavesdropper. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide an optimization method for the coordinated secure communication network of offshore distributed distribution units and drones. By introducing the collaborative architecture of the control center, multi-antenna drone relay and single-antenna distributed distribution units, a set of power optimization algorithms and trajectory optimization algorithms are designed to maximize the safety and reliability of the distribution unit dispatching system while improving the communication rate, meeting the distribution network operation and management requirements for safe access of large-scale distributed energy at sea.

[0006] Technical solution: The present invention provides an optimization method for an offshore distributed power distribution unit and a drone coordinated secure communication network, wherein the communication network hardware includes an offshore distributed power distribution unit, a drone relay, and a control center, and the method includes the following steps:

[0007] Step 1: Modeling multiple communication link channels from the control center to the distributed distribution unit;

[0008] Step 2: Model the signal transmission strategy between the control center and the drone relay;

[0009] Step 3: Model the signal eavesdropping risk link and evaluate the eavesdropping channel capacity of the two scheduling links from the control center to the potential eavesdropper and from the drone relay to the potential eavesdropper.

[0010] Step 4: Formulate information security issues;

[0011] Step 5: Design of optimization scheme for scheduling problem;

[0012] Step 6: Output the optimized distribution network dispatching parameters, including the transmission power matrix, UAV trajectory and distribution unit grouping parameters.

[0013] Furthermore, in step 1, the multi-communication link channel from the control center to the distributed power distribution unit is modeled; specifically comprising the following steps:

[0014] Step 1.1, model the direct transmission link from the control center to the distributed distribution unit, and use the Alpha-beta-gamma channel model ABG to approximate the path loss of actual communication;

[0015] Step 1.2: Model the transmission link from the control center to the relay UAV, using a small-scale Rayleigh fading model. Considering that the UAV is affected by sea waves, the line-of-sight component and the non-line-of-sight component coexist in the model.

[0016] Step 1.3: The link from the relay UAV to the distributed distribution unit also uses the coexistence of line-of-sight components and non-line-of-sight components, and the channel form is a dual form.

[0017] Furthermore, in step 2, the signal transmission strategy modeling for the control center and the drone relay includes the following steps:

[0018] Step 2.1, the control center directly schedules link modeling, including the transmission signal model, the receiving signal model and the channel capacity evaluation;

[0019] Step 2.2, UAV relay scheduling link modeling, including the signal model sent from the control center to the UAV relay, the signal model sent from the UAV relay to the distribution unit, and the scheduling link channel capacity evaluation method finally accepted by the distribution unit.

[0020] Furthermore, in step 4, the information security problem is formulated, which specifically includes the following steps:

[0021] Step 4.1, model the confidentiality communication of the links from the control center to the distributed distribution unit and from the drone relay to the distributed distribution unit, and give the average confidentiality capacity expression for multiple time slots;

[0022] In step 4.2, the optimization problem is decomposed to decompose the problem of maximizing the confidentiality capacity into maximizing the confidentiality capacity of the direct scheduling link and maximizing the confidentiality capacity of the relay scheduling link.

[0023] Furthermore, in step 5, the scheduling problem optimization solution design specifically includes the following steps:

[0024] Step 5.1, using K-means clustering algorithm to group distributed distribution units;

[0025] Step 5.2: Optimize the power of the direct scheduling link based on the zero forcing method;

[0026] Step 5.3, optimize the UAV relay scheduling link power;

[0027] Step 5.4: Use the deep Q network to optimize the UAV flight trajectory.

[0028] Furthermore, step 5.1 specifically includes the following steps:

[0029] Step 5.1.1, initialize the cluster center of the K-means clustering algorithm;

[0030] Step 5.1.2, calculate the distance between each data point and the cluster center, and assign the data point to the cluster center;

[0031] Step 5.1.3, update the centroid of the K-means clustering algorithm and repeat the iteration;

[0032] Step 5.1.4: Use the distributed distribution unit clustering algorithm to group the distribution units.

[0033] Furthermore, step 5.3 specifically includes the following steps:

[0034] Step 5.3.1, decompose the channel matrix based on generalized singular value decomposition and transform the original problem form;

[0035] Step 5.3.2: Based on the transformed power optimization problem, construct the Lagrangian function and KKT conditions and solve them;

[0036] Step 5.3.3: Based on the optimal solution of the Langladi function, the sub-channel auxiliary parameters are determined to determine the optimal transmission power of the sub-channel.

[0037] Furthermore, step 5.4 specifically includes the following steps:

[0038] Step 5.4.1. Construct a reward function based on the UAV motion pattern and scheduling problem.

[0039] Step 5.4.2, obtain the supervised learning parameters of the deep Q network by calculating the Bellman equation;

[0040] Step 5.4.3, design the loss function for deep Q network training;

[0041] Step 5.4.4, determine the loss gradient during the deep Q network training process based on the loss function;

[0042] Step 5.4.5: Use the UAV trajectory optimization algorithm of the distributed distribution unit dispatching system to optimize the UAV trajectory.

[0043] The present invention discloses an offshore distributed power distribution unit-UAV collaborative safety communication network, wherein the hardware body comprises a control center with multiple antennas, a multi-antenna UAV dispatching relay and multiple single-antenna offshore distributed power distribution units; the software body comprises a dispatching and power optimization unit located in the control center, and a power optimization unit and a trajectory optimization unit located in the UAV.

[0044] Furthermore, the control center equipped with M antennas communicates control signals with K single-antenna distributed power distribution units at the same frequency through direct communication or by using drones equipped with N antennas. The signal transmission adopts space division multiple access and time division multiple access technology. The control center directly sends control signals to K b The distribution terminal transmits control instructions and relays them to K r The distribution terminals are controlled, and K = K b +K r .

[0045] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0046] (1) The present invention proposes a UAV cooperative communication network for offshore distributed distribution units. By introducing a collaborative architecture of a control center, a multi-antenna UAV relay, and a single-antenna distributed distribution unit, a set of power optimization algorithms and trajectory optimization algorithms are designed to maximize the safety and reliability of the distribution unit dispatching system while improving the communication rate, thus meeting the distribution network operation and management requirements for safe access of large-scale distributed energy at sea.

[0047] (2) The present invention solves the problems of limited communication coverage, insufficient real-time dispatching capability and weak communication security in large-scale distributed energy access at sea, improves the operating efficiency and security of the distribution network, and meets the needs for efficient communication and safe dispatching in new energy access scenarios.

[0048] (3) The present invention optimizes the secure communication network by combining convex optimization with machine learning. While ensuring good optimization effects, it does not occupy high computing power, effectively saving the computing power resources and costs of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of the coordinated secure communication network for offshore distributed power distribution units and drones;

[0050] Figure 2 Flowchart of the method for optimizing the collaborative secure communication network of UAVs for offshore distributed power distribution units;

[0051] Figure 3 This is a comparison chart of the maximum transmission rate in different scenarios;

[0052] Figure 4 The following is a comparison chart of the maximum confidentiality transmission rate in different scenarios. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0054] like Figure 1 The figure shows a schematic diagram of the coordinated safety communication network of the offshore distributed power distribution unit and the drone. The control center equipped with M antennas can communicate control signals with K single-antenna distributed power distribution units at the same frequency through direct communication or by using drone relays equipped with N antennas. The signal transmission adopts space division multiple access and time division multiple access technology. The control center directly sends K b The distribution terminal transmits control instructions and relays them to K r The distribution terminals are controlled, and K = K b +K r .

[0055] In order to fully convey the dispatch signal, the transmission rate of the control center and the drone relay needs to be maximized. At the same time, in order to ensure the security of transmission, that is, there may be potential eavesdroppers eavesdropping on the dispatch signal, thereby creating a security risk for the communication system, the present invention designs a power optimization algorithm and a trajectory optimization algorithm to maximize the security of the communication network while meeting the dispatch link communication rate requirements.

[0056] In order to maximize the secure transmission rate of the communication network, it is necessary to optimize the scheduling parameters, such as Figure 2 The figure shows a flow chart of the optimization method of the cooperative safety communication network of the UAV of the offshore distributed power distribution unit. The technical solution of the invention is described in detail below in conjunction with the accompanying drawings:

[0057] 1. Multi-communication link channel modeling

[0058] (1) Control center-distributed distribution unit link

[0059] The ground-to-ground communication channel uses the Alpha-beta-gamma (ABG) channel model to describe the communication channels between the control center and potential eavesdroppers and between distributed distribution units and potential eavesdroppers. This model is the closest approximation model to the path loss of actual 5G ground communication measurement results and is adopted by standardization organizations such as ITU-R, 3GPP, mmMAGIC and QuaDRiGa. It is defined as:

[0060]

[0061] Among them, d gg represents the two-dimensional distance between the transmitting node and the receiving node, is the intercept, ρ G and γ G The shadow fading coefficient χ G2G,σ is modeled as having a mean of zero and a standard deviation of σ sh is a Gaussian random variable.

[0062] (2) Control center-relay drone link

[0063] UAV-to-PDU communication link: When modeling the air-to-ground (A2G) communication channel between the UAV and the ground receiver, the present invention adopts a small-scale Rayleigh fading model (Rician Fading), in which the line-of-sight (LoS) component and the non-line-of-sight (NLoS) component coexist. The ground control center and the UAV receiver (UAV relay) are both equipped with a uniform linear array (ULA) consisting of M and N antennas, respectively. The channel model between the control center and the relay UAV is a composite channel with LoS (line-of-sight) and NLoS (non-line-of-sight) superimposed, and its path loss is expressed as follows:

[0064]

[0065] Where λ0 represents the path loss per 1 meter; d TR is the three-dimensional distance between the control center and the relay UAV; α is the path loss exponent; β is the Rayleigh fading factor.

[0066] Assume that the non-line-of-sight channel component The elements of are independent and identically distributed and follow the distribution Line-of-sight channel component It can be expressed as the arrival angle component G AoA and the deviation angle component G AoD The product of, that is:

[0067]

[0068] Among them, the arrival angle component G between the control center and the relay drone is AoA and the deviation angle component G AoD Can be expressed as and The parameter λ is the carrier wavelength, Y is the antenna spacing, and Λ TR =cosΘsinφ are the components of the angle of arrival, Θ is the azimuth angle, φ is the elevation angle, is the deviation angle component, where is the pitch angle, and ψ is the azimuth angle. These parameters are used to describe the signal propagation characteristics between the control center and the relay drone.

[0069] (3) Relay UAV-Distributed Power Distribution Unit Link

[0070] Considering a single antenna setting at each distributed power distribution unit, the transmission path loss from the relay drone to the distributed power distribution unit is expressed as follows:

[0071]

[0072] Among them, the non-line-of-sight channel component Follow the distribution Line-of-sight channel component The expression of is similar to formula (3).

[0073] 2. Signal transmission strategy modeling

[0074] The control center equipped with M antennas can communicate control signals with K single-antenna distributed power distribution units at the same frequency through direct communication or by using drones equipped with N antennas to relay. The signal transmission adopts space division multiple access (SDMA) and time division multiple access (TDMA) technology. b The distribution terminal transmits control instructions and relays them to K r The distribution terminals are controlled, and K = K b +K r The following will provide a detailed description of the corresponding control signal transmission link model and transmission rate.

[0075] 1) Control center direct scheduling link

[0076] For direct communication, the control center performs beamforming design to the distributed distribution terminals based on space division multiple access and time division multiple access modes. The transmit beam assigns a beam vector to each distribution unit. However, transmit power leakage may occur between different beams, resulting in mutual interference between multiple distribution terminals. The present invention considers the downlink transmission scenario, in which the control center sends K b Distributed distribution units transmit K b data streams. Its transmission signal model is:

[0077]

[0078] in is the beamforming vector, s k is the information symbol transmitted for the kth distribution unit. The beamforming or precoding matrix of the control center contains K b beamforming vectors, in Then, by the beamforming vector ||w b,k || 2 The square norm of is used to calculate the transmission power allocated to the kth power distribution unit. b The received signal of each distribution unit is expressed as

[0079]

[0080] For the kth distribution unit, the first element on the right side of the above formula (6) is the channel coefficient h from the distribution unit to the control center 0,k , beamforming vector w b,k and information symbols k The rest, relative to the control signal s k , are all interference. Including sending to the remaining K b -1 device’s signal, represented by the beamforming vector w b,i and information i Composition, and Gaussian noise in the environment Then, we get the signal-to-interference-noise ratio

[0081]

[0082] in, represents the MISO channel from the control center to the kth distribution unit. The channel capacity of the direct link is obtained by the following formula:

[0083] C b,k =log2(1+γ b,k ). (8)

[0084] (2) UAV relay dispatch link

[0085] Assume time slot-based synchronization between the control center transmission and the drone relay transmission. In odd time slots (phases), the control center sends K r The data streams are transmitted to the drone relay, and each data stream corresponds to a distribution unit. The transmission between the control center and the drone relay can be modeled as a standard point-to-point MIMO channel. The signal received by the drone relay can be expressed as:

[0086]

[0087] in, is the signal vector received by the drone relay; It is the MIMO communication channel between the control center and the drone relay; is a noise vector, which conforms to the complex Gaussian distribution, has a mean of zero, and a covariance matrix that is the identity matrix; is the signal sent by the control center, where w b,k is the beam vector of the kth distribution unit, s k is the data flow of the kth power distribution unit.

[0088] In even time slots, the drone relays the transmission signal:

[0089] x r =W r y1, (10)

[0090] in, The transmission signal relayed by the drone, is the beamforming matrix. K r The signal received by a power distribution unit can be expressed as:

[0091]

[0092] in, is the signal vector received by the power distribution unit, It is a drone relay and K r A2G communication channel between the power distribution units, is a noise vector, which conforms to a complex Gaussian distribution, has a mean of zero, and a covariance matrix that is the identity matrix.

[0093] The signal received by the kth power distribution unit is:

[0094]

[0095] in, represents the MISO channel relayed from the UAV to the kth PDU, is additive noise.

[0096] The signal-to-noise ratio (SINR) of the communication link from the control center to the kth distribution unit through the drone relay can be calculated as:

[0097]

[0098] The channel capacity of the indirect link C r,k It can be obtained by formula (8), where γ r,k Replace γ b,k It is important to note that H1 and H2 are directly affected by the motion of the UAV (e.g., changes in distance, altitude, and orientation relative to the ground receiver).

[0099] 3. Signal eavesdropping risk link modeling

[0100] (1) Direct scheduling link eavesdropping

[0101] The eavesdropper monitors the direct link between the control center and the relevant distribution unit, and the received signal is:

[0102]

[0103] in, represents the communication channel between the control center and the eavesdropper, n e is the noise at the eavesdropper, satisfying Based on the beamforming of the distribution unit k, the SINR of the direct link between the control center and the eavesdropper can be expressed as

[0104]

[0105] Therefore, the direct link eavesdropping capacity from the control center to the kth distribution unit is

[0106]

[0107] (2) Relay dispatch link eavesdropping: The eavesdropper can eavesdrop on the A2G relay communication link between the drone relay and the power distribution unit. The received signal can be expressed as

[0108]

[0109] The capacity of the eavesdropper-associated relay link is expressed as

[0110]

[0111] Among them, γ r,e is SINR, represents the channel between the drone and the eavesdropper, n 2,e is the noise at the eavesdropper, satisfying

[0112] 4. Formulate information security issues

[0113] Confidentiality capacity is a measure of the rate at which information can be transmitted in a secure manner without being eavesdropped. It is defined as the difference between the achievable data rate of the legitimate receiver and the achievable data rate of the eavesdropper, taking into account the channel conditions and the security measures adopted. Confidentiality capacity corresponds to the rate at which the eavesdropper cannot decode any data. For the system model in Section 1, K scheduled directly by the control center b The average total confidentiality capacity of a distributed distribution unit in T time slots is

[0114]

[0115] Similarly, K dispatched by drone relay r The average total confidentiality capacity of a distributed distribution unit in T time slots can be expressed as

[0116]

[0117] Among them, the symbol [·] + The meaning of is to take a non-zero value, that is, The total confidentiality capacity is the sum of the capacity of the direct transmission link and the capacity of the drone relay transmission link, which can be expressed as

[0118] C T =C sec,b +C sec,r (twenty one)

[0119] The above formulas (19)-(21) characterize the quantitative information security indicators of the distributed distribution network dispatching network, the purpose of which is to minimize the channel capacity of the eavesdropper on the basis of maximizing the confidentiality capacity.

[0120] Considering the two scheduling modes of directly scheduling links or scheduling distribution units through drones, the present invention decomposes the problem into two optimization problems:

[0121] 1) Direct dispatch link: For the directly dispatched distribution unit, the optimization problem can be expressed as

[0122]

[0123] In the above formula, C sec,b is the confidentiality capacity defined by equation (19), P b,max is the maximum dispatch signal transmission power of the control center, P b is the actual dispatch signal transmission power of the dispatch center, which can be expressed as The superscript of is the conjugate transpose. The optimization goal of the above problem (22) requires understanding the situation of the eavesdropping channel, but this is an unrealistic assumption in the actual communication environment. Therefore, the present invention assumes that the location of the eavesdropper and its channel state information (CSI) are unknown, which is a common situation in practice. In this case, due to the passive characteristics of the eavesdropper, it is difficult to detect or estimate its existence, location or channel. Therefore, the present invention can only consider the capacity of the legitimate distribution unit and reformulate the optimization problem as follows:

[0124]

[0125] The eavesdropper's location and channel are only used to calculate the resulting confidentiality capacity for performance evaluation.

[0126] 2) UAV relay scheduling link: For distributed power distribution units scheduled via UAV relay, since the location and channel state information (CSI) of the eavesdropper are unknown, the optimization problem can be defined as

[0127]

[0128] Among them, the objective function is the confidentiality capacity, P r,max is the maximum transmit power of the drone, is the actual transmission power of the UAV, (x r ,y r ,z r ) is bound to the spatial domain (L x ,L y ,L z )’s three-dimensional coordinates of the drone.

[0129] 5. Design of optimization scheme for scheduling problem

[0130] In order to solve the above scheduling optimization problem, the present invention proposes an optimization method, which decomposes the original problem into a direct scheduling link optimization problem and a UAV scheduling link optimization problem. The direct scheduling link optimization problem is characterized by the beamforming problem of the scheduling signal, and the optimization problem of the UAV scheduling link is characterized by the beamforming problem and trajectory optimization problem of the UAV. The specific solution steps are as follows:

[0131] (a) Distribution unit grouping

[0132] The goal of distribution unit grouping is to divide K distribution units into two groups, one of which is directly scheduled by the control center and the other is scheduled by the drone relay.

[0133] In order to solve the problem of grouping distribution units, an exhaustive search method can be used, but this method has a high computational complexity and grows exponentially with the increase in the number of distribution units. Therefore, the present invention adopts K-means clustering grouping method, which is an unsupervised machine learning algorithm used to group a group of objects so as to maximize the similarity of members within the group and maximize the difference with other group members. K-means clustering is applicable to any single-dimensional or multi-dimensional data measurement, and can be grouped according to the target number of clusters defined by the distribution unit. Compared with graph theory, fuzzy c-means clustering, and hierarchical clustering, K-means clustering has higher computational efficiency.

[0134] The present invention combines the characteristics of wireless communication systems to determine the similarity of data points. Since the goal is to assign distribution units to different scheduling links, one is a direct scheduling link and the other is a relay scheduling link, the present invention uses the normalized channel coefficients between distributed distribution units and the control center as data points for the K-means clustering algorithm. This method captures the channel gain changes caused by RF propagation effects (such as small-scale fading and shadow fading). Therefore, it can be defined

[0135]

[0136] in, is the normalized channel gain, h b,k is the channel gain between the control center and the kth distribution unit device, and |.|2 is the L2 vector norm. After using these channel gains as data points, the present invention applies the K-means clustering algorithm to determine the cluster center or centroid, and further determines the distribution unit associated with each centroid. The goal is to use the similarity of the channel between the control center and the distributed distribution units to divide the distributed distribution units into two groups, one group is directly dispatched by the control center, and the other group is dispatched by the drone relay. The specific steps involved are as follows:

[0137] (a1) Initialize the centroid: Randomly select the initial centroid C = {c1, c2, ... c n}, as K available data points U = {u1,u2,...,u k ...,u K In this invention, two clusters are considered: c1 corresponds to the direct scheduling link, c2 corresponds to the drone relay scheduling link, and among the K distributed distribution units, the data point of the kth distribution unit is

[0138] (a2) Assigning data points: Calculate the distance between each data point (such as channel state) and the cluster center, and assign the data point to the nearest cluster center. A variety of metrics can be used to calculate the distance between data points, such as Euclidean distance, Manhattan distance, etc. The present invention uses the L2 norm (i.e., Euclidean distance).

[0139] (a3) Update the centroid: Update the centroid to minimize the sum of the squared distances between the distribution unit and its centroid, Among them, d r,k =|u k -c r |2 2 is the centroid The Euclidean distance between the normalized channel gain and the centroid is

[0140] Combining the above steps, a distributed distribution unit clustering algorithm is given. The specific process is as follows:

[0141] Algorithm 1: Distributed Distribution Unit Clustering

[0142] Input: U and C

[0143] Output: K c and C,

[0144] 1. Initialize the cluster center set: And set c=1;

[0145] 2. While loop: When c ≤ C:

[0146] a. Randomly select a cluster center from U

[0147] b. Update

[0148] c. Set c = c + 1;

[0149] 3. End While

[0150] 4. Repeat the cycle:

[0151] a. For each distribution unit m∈U, calculate its The minimum distance;

[0152] b. Assign each distribution unit to the nearest cluster center;

[0153] c. Update cluster centers is the average value of all distribution units m;

[0154] 5. Until the cluster distribution unit no longer changes.

[0155] (b) Direct scheduling link power optimization problem

[0156] Problem (23) is a traditional beamforming power control problem involving a control center and a distribution unit, which has been widely studied. The beamforming vector of the control center can be obtained by applying the weighted minimum mean square error (WMMSE) algorithm, which is an iterative closed-form solution that optimizes the precoding vectors of the transmitter and receiver to maximize the total dispatch rate of all distribution units under the power constraint of the control center. The precoding solution for the direct communication link is

[0157]

[0158] In the above formula, Q = diag{q1,...,q k} is the receiving end precoding matrix, F = diag{f1,...,f k} is the weight matrix, I M is the covariance matrix.

[0159] (c) UAV relay scheduling link power optimization problem

[0160] Next, the present invention will solve the beamforming and power control problems in relay communication. Inspired by the ZeroForing criterion and the channel singular value decomposition (SVD) structure, the present invention first proposes a beamforming matrix structure (i.e., W r ) to eliminate the interference between the dispatching signals of the distribution units. This transforms the optimization problem (24) into a simplified convex optimization problem. Then, the present invention uses the Lagrangian function and the KKT condition to solve the modified optimization problem to obtain the optimal beamforming matrix of the UAV.

[0161] 1. Beamforming matrix design: Beamforming is performed at the dispatch center and the drone relay. Each beamforming matrix provides a dispatch signal for a group of distribution units. The dispatch signal sent by the drone relay to the $K$th distribution unit can be expressed as:

[0162] y2=H2W r H1W br s K +H2W r n1+n2 (27)

[0163] In the above formula, represents K groups of signals sent to K distribution units. The zero-forcing criterion requires H2W r H1W bris a diagonal matrix of rank K, which means Rank(H1) ≥ K and Rank(H2) ≥ K. Based on the generalized singular value decomposition (GSVD) theory, H2 and H1 can be expressed as:

[0164]

[0165] Among them, U i and V i (for i=1,2) is the identity matrix, Is a diagonal matrix with positive diagonal elements. Knowing the channel coefficients of the control center and the drone, in order to satisfy the ZF criterion, the beamforming matrix W of the direct scheduling link is br for The beamforming matrix W of the UAV relay scheduling link r for Among them, Λ b , and Λ r Both are K×K diagonal matrices. To simplify the problem, it can be assumed that the elements of these two diagonal matrices are non-negative, representing the beamforming power allocated by the control center and the UAV relay, respectively.

[0166] Based on the above decomposition method, the objective function defined in formula (24) can be written as

[0167]

[0168] Among them, λ r,k Yes r (k,k). Furthermore, the beamforming power constraint defined in equation (24) can be expressed as

[0169]

[0170] Among them, σ 2,n is Σ2(n,n). Combining equations (29) and (30), the beamforming power optimization problem of the UAV at any location can be written as:

[0171]

[0172] Among them, P r,max is the maximum available transmission power of the UAV, λ r,max is the maximum allocated power of each antenna. The Lagrangian function of the optimization problem can be expressed as:

[0173]

[0174] Among them, α1,α 2,l and α 3,lis the non-negative Lagrange multiplier corresponding to the first and second constraints. It is known that the drone antenna array has a total of N antennas, and it is necessary to design a beamforming power matrix for each sub-antenna. Define an auxiliary parameter as

[0175]

[0176] Assume that the lth antenna meets the condition Then the power matrix of the lth antenna is 0. Assume that the lth antenna satisfies the condition Then the power matrix of the lth antenna is Assume that the lth antenna meets the condition Then the power matrix of the lth antenna is λ r,max Among them, the Lagrange multiplier It can be obtained by substituting (33) into the first constraint of (31) and making it hold. The optimal beamforming matrix and transmission power formula defined above will be used for UAV trajectory optimization.

[0177] (d) UAV relay flight trajectory optimization problem

[0178] The objective function (24) is the parameter x r ,y r ,z r ,P and the constraints are non-convex, and the problem is a non-deterministic polynomial. Therefore, the present invention proposes a machine learning solution for updating the trajectory of the drone based on the transition process of the current system state. Since the next system state is independent of the previous state and action, the process can be modeled as a Markov decision process. In order to avoid the computational difficulties caused by the high-dimensional state-action space, the present invention proposes a deep Q network. It should be noted that the deep Q network proposed below is based on basic reinforcement learning algorithms such as Q-learning and deep reinforcement learning. The purpose is to replace traditional optimization tools with deep Q networks to solve this non-deterministic polynomial optimization problem while reducing power consumption and computing resource consumption. Depending on the specific application, the framework can be expanded to a more advanced learning model to adapt to specific usage scenarios.

[0179] The Markov decision process of the drone agent consists of the state space Action Space Reward Space and the transition probability space At time slot t, the agent observes the state and take actions according to its strategy According to the transfer probability distribution, the agent will move to the new state s t+1Since the transition probabilities are specific to the operating environment, the present invention chooses the Q-learning method as a model-free algorithm to find the best strategy for each action in each state. This means that there is no need to know But states, actions, and rewards need to be defined carefully.

[0180] The state set of this network is defined as Where t is the time slot index. Each state s t The corresponding 3D coordinates of the drone and the distribution unit that the drone relay is responsible for dispatching. The drone status is based on the defined action set transfer, where each action consists of three parts related to the movement of the drone, namely a t ={δ x ,δ y ,δ z}, where δ x , δ y and δ z Indicates taking action a in the x, y, z directions t After that, the drone relay will receive a reward R t (s t ,a t ). The drone will receive more rewards for actions that lead to higher safe scheduling link capacity. The reward function is defined as:

[0181]

[0182] The deep Q network was originally proposed by Google Deep Mind and combines reinforcement learning and deep learning methods. This technology uses nonlinear functions, especially deep neural networks (DNNs), to approximate Q values ​​and handle high-dimensional state-action problems. The deep Q network uses two deep neural networks with the same structure, including a training network and a target network. The training network outputs the Q value associated with the drone's action in each state, and the target network obtains the target Q value by calculating the Bellman equation, thereby supervising the training network. The Bellman equation it calculates is expressed as

[0183]

[0184] Where s and a represent the next state and action, respectively. The parameter γ∈(0,1) is a discount factor that indicates the importance of future rewards. The target value is compared with the output of the training network to minimize the loss function:

[0185]

[0186] The Q value of the first term comes from the target network, and the Q value of the second term comes from the training network. and θ represent the weights of the target network and the training network, respectively. The coefficients of are updated every few time steps to ensure the stability of the target value, thus promoting stable learning.

[0187] When the drone flies, the system generates an experience record. At time step t, the experience contains the current state s t 、Action a t , Reward t and the next state s t+1 , forming a four-tuple e t =(s t ,a t ,r t ,s t+1 ). Each experience is stored in a replay memory with a capacity of N, denoted as M = {e1,...,e t ,...,e N}. The memory is a queue-like buffer that stores the latest N experience vectors. A small batch of samples is used from the replay memory as input for training the network. The main reason for using a small batch of samples is to break the sequential correlation of the environment state, thereby promoting generalization.

[0188] In the training process of neural network, loss gradient is needed, which is mathematically expressed as

[0189]

[0190] In order to update the weight parameters θ and The goal is to minimize the prediction error. Finally, the present invention adopts an ∈-greedy algorithm to select actions to balance the exploration and exploitation of the drone in the environment. In this algorithm, the drone explores the environment by selecting random actions with a probability of ∈. More specifically, the drone exploits the environment by selecting actions that maximize the Q-value function with a probability of 1-∈, i.e. Initially, ∈ is set to a high value so that the drone spends more time exploring. As the agent gradually gains more information about the environment, the value of ∈ is gradually reduced to make full use of existing experience and select the best action instead of continuing to explore a lot.

[0191] Combining the above process and formula, a UAV trajectory optimization algorithm for distributed distribution unit dispatching system is proposed. The specific process is as follows:

[0192] Algorithm 2: UAV trajectory optimization algorithm for distributed distribution unit dispatching system

[0193] 1. Initialize the exploration rate ε start , ε end and decay rate decay

[0194] 2. Initialize T time steps, Rounds

[0195] 3. Initialize playback memory M, capacity N

[0196] 4. Initialize θ, γ,α,B

[0197] Start loop Each round performs the following operations:

[0198] Reset Environment

[0199] Start the loop t=1,2,...,T and perform the following operations

[0200] Get the initial observation state s t

[0201] If ∈>random(0,1)then

[0202] Randomly select action a from the action space A t

[0203] otherwise

[0204] Select action a t = argmax a∈A Q(s t ,a;θ)

[0205] End sub-judgment

[0206] The drone performs actions in the environment t ;

[0207] Observe the state transition s t+1 ,r t ;

[0208] The storage state is transferred to M: M←M∪{s t ,a t ,r t ,s t+1};

[0209] Randomly sample a mini-batch of samples from M: e i =(s i ,a i ,r i ,s i+1 ); train the deep neural network and calculate the estimated Q value;

[0210] Calculate the loss between the estimated Q value and the calculated Q value;

[0211] Derive the loss gradient and update the parameters θ in the training network;

[0212] Every B steps, copy θ to Update the target network;

[0213] End sub-loop

[0214] Update ∈ via ∈-greedy algorithm

[0215] Store the rewards for each round

[0216] End loop output: optimized drone trajectory.

[0217] like Figure 3 The maximum transmission rate comparison chart under different scenarios shows the maximum transmission rate of drone trajectory optimization combined with beamforming optimization under different numbers of distribution units. As can be seen from the figure, as the number of distribution units increases, the maximum transmission rate increases accordingly. The proposed drone trajectory optimization strategy and power optimization strategy can significantly improve the information transmission rate of the control center.

[0218] like Figure 4 The maximum confidentiality transmission rate comparison chart under different scenarios shows the maximum confidentiality transmission rate under different numbers of power distribution units when the drone trajectory optimization is combined with beamforming optimization. As can be seen from the figure, the proposed drone trajectory optimization strategy and power optimization strategy can significantly improve the security transmission rate of the target control center, especially when comparing the two scenarios of only trajectory optimization and no optimization, the security rate is significantly improved.

Claims

1. A method for optimizing an offshore distributed power distribution unit UAV collaborative safety communication network, characterized in that: The communication network hardware includes an offshore distributed power distribution unit, a drone relay and a control center, and the method includes the following steps: Step 1: Modeling multiple communication link channels from the control center to the distributed distribution unit; Step 2: Model the signal transmission strategy between the control center and the drone relay; Step 3: Model the signal eavesdropping risk link and evaluate the eavesdropping channel capacity of the two scheduling links from the control center to the potential eavesdropper and from the drone relay to the potential eavesdropper. Step 4: Formulate information security issues; Step 5: Design of optimization scheme for scheduling problem; Step 6: Output the optimized distribution network dispatching parameters, including the transmission power matrix, UAV trajectory and distribution unit grouping parameters.

2. The optimization method of the offshore distributed power distribution unit UAV collaborative safety communication network according to claim 1 is characterized in that: In step 1, the modeling of multiple communication link channels from the control center to the distributed power distribution unit specifically includes the following steps: Step 1.1, model the direct transmission link from the control center to the distributed distribution unit, and use the Alpha-beta-gamma channel model ABG to approximate the path loss of actual communication; Step 1.2: Model the transmission link from the control center to the relay UAV, using a small-scale Rayleigh fading model. Considering that the UAV is affected by sea waves, the line-of-sight component and the non-line-of-sight component coexist in the model. Step 1.3: The link from the relay UAV to the distributed distribution unit also uses the coexistence of line-of-sight components and non-line-of-sight components, and the channel form is a dual form.

3. The optimization method of the offshore distributed power distribution unit UAV collaborative safety communication network according to claim 1 is characterized in that: In step 2, the signal transmission strategy modeling for the control center and the drone relay includes the following steps: Step 2.1, the control center directly schedules link modeling, including the transmission signal model, the receiving signal model and the channel capacity evaluation; Step 2.2, UAV relay scheduling link modeling, including the signal model sent from the control center to the UAV relay, the signal model sent from the UAV relay to the distribution unit, and the scheduling link channel capacity evaluation method finally accepted by the distribution unit.

4. The optimization method of the offshore distributed power distribution unit UAV collaborative safety communication network according to claim 1 is characterized in that: In step 4, the information security problem is formulated, which specifically includes the following steps: Step 4.1, model the confidentiality communication of the links from the control center to the distributed distribution unit and from the drone relay to the distributed distribution unit, and give the average confidentiality capacity expression for multiple time slots; In step 4.2, the optimization problem is decomposed to decompose the problem of maximizing the confidentiality capacity into maximizing the confidentiality capacity of the direct scheduling link and maximizing the confidentiality capacity of the relay scheduling link.

5. The optimization method of the offshore distributed power distribution unit UAV collaborative safety communication network according to claim 1 is characterized in that: In step 5, the optimization scheme for the scheduling problem is designed, which specifically includes the following steps: Step 5.1, using K-means clustering algorithm to group distributed distribution units; Step 5.2: Optimize the power of the direct scheduling link based on the zero forcing method; Step 5.3, optimize the UAV relay scheduling link power; Step 5.4: Use deep Q network to optimize the UAV flight trajectory.

6. The optimization method of the offshore distributed power distribution unit UAV collaborative safety communication network according to claim 5 is characterized in that: Step 5.1 specifically includes the following steps: Step 5.1.1, initialize the cluster center of the K-means clustering algorithm; Step 5.1.2, calculate the distance between each data point and the cluster center, and assign the data point to the cluster center; Step 5.1.3, update the centroid of the K-means clustering algorithm and repeat the iteration; Step 5.1.4: Use the distributed distribution unit clustering algorithm to group the distribution units.

7. The optimization method of the offshore distributed power distribution unit UAV collaborative safety communication network according to claim 5 is characterized in that: Step 5.3 specifically includes the following steps: Step 5.3.1, decompose the channel matrix based on generalized singular value decomposition and transform the original problem form; Step 5.3.2: Based on the transformed power optimization problem, construct the Lagrangian function and KKT conditions and solve them; Step 5.3.3: Based on the optimal solution of the Langladi function, the sub-channel auxiliary parameters are determined to determine the optimal transmission power of the sub-channel.

8. The optimization method of the offshore distributed power distribution unit UAV collaborative safety communication network according to claim 5 is characterized in that: Step 5.4 specifically includes the following steps: Step 5.4.

1. Construct a reward function based on the UAV motion pattern and scheduling problem. Step 5.4.2, obtain the supervised learning parameters of the deep Q network by calculating the Bellman equation; Step 5.4.3, design the loss function for deep Q network training; Step 5.4.4, determine the loss gradient during the deep Q network training process based on the loss function; Step 5.4.5: Use the UAV trajectory optimization algorithm of the distributed distribution unit dispatching system to optimize the UAV trajectory.

9. A marine distributed distribution unit drone collaborative safety communication network, used to implement the method according to claim 1, characterized in that: The hardware body includes a control center with multiple antennas, a multi-antenna UAV dispatching relay and multiple single-antenna offshore distributed power distribution units; the software body includes a dispatching and power optimization unit located in the control center, as well as a power optimization unit and trajectory optimization unit located in the UAV.

10. The offshore distributed power distribution unit-drone collaborative safety communication network according to claim 9, characterized in that: The control center equipped with M antennas communicates control signals with K single-antenna distributed power distribution units at the same frequency through direct communication or by using drones equipped with N antennas. The signal transmission adopts space division multiple access and time division multiple access technology. The control center directly sends control signals to K b The distribution terminal transmits control instructions and relays them to K r The distribution terminals are controlled, and K = K b +K r .

Images