Available link intelligent prediction and optimal selection method for unmanned cluster anti-interference communication

Abstract

The invention relates to an intelligent available link prediction and optimal selection method for unmanned cluster anti-interference communication. The method comprises the following steps: acquiring a multi-channel interference power matrix; an LSTM-KAN interference prediction model is established; predicting channel interference power of a plurality of time slots in the future by using an LSTM-KAN interference prediction model; channel selection is modeled as a dual-objective optimization model for minimizing channel switching times and minimizing channel average interference power; and solving the dual-objective optimization model based on a classification sorting genetic algorithm, improving a crossover variation strategy and an elitist retention strategy to increase the generation probability of a low-channel switching frequency solution, and retaining the low-channel switching frequency solution as much as possible in an iteration process, and finally generating an optimized channel switching and interference power solution. Compared with the prior art, the method has the advantages of remarkably reducing the channel switching times, improving the throughput and reducing the packet loss rate, and has important application value in resisting unmanned trunking communication in a strong electromagnetic interference environment.

Description

Technical Field

[0001] The present invention relates to the field of unmanned cluster anti-interference communication, and in particular to an available link intelligent prediction and optimal selection method for unmanned cluster anti-interference communication. Background Art

[0002] In recent years, unmanned swarms have been widely used in various tasks such as reconnaissance, attack, and rescue. The communication network that connects multiple platforms is the basis for unmanned swarms to perform tasks in a coordinated manner. However, in a confrontational environment, unmanned swarms often face strong electromagnetic interference, requiring the communication network to have strong interference resistance. In addition to the commonly used frequency hopping, spread spectrum, and directional antenna anti-interference methods, real-time perception and prediction of channel status and optimization of working channels is an important anti-interference method.

[0003] According to the different prediction contents, interference channel prediction can be divided into two categories: channel occupancy prediction and channel interference power prediction. Channel occupancy prediction is a binary classification prediction of the presence or absence of channel interference. The prediction methods include regression analysis, hidden Markov model and neural network. The regression analysis method predicts by finding the functional relationship between time series, but it is difficult to handle the nonlinear relationship caused by complex interference; the hidden Markov model method predicts the next state by using the state transition matrix, but the single-step dependency characteristic leads to low prediction accuracy; the prediction methods based on neural networks such as MLP and LSTM have become important technologies for channel situation prediction, but they require sufficient data sets for training. Channel interference power prediction is a regression prediction of the interference power value perceived by the spectrum. The prediction method is mainly a neural network method, such as using the LSTM network to realize power spectrum density prediction and randomly select channel access below the interference power threshold, and using the LSTM network to realize channel power prediction and select the channel access with the maximum expected throughput.

[0004] Channel occupancy prediction depends on the accuracy of spectrum sensing judgment, and the binary classification characteristics mean that channel selection can only randomly select interference-free channels. Channel interference power prediction can reflect the interference changes caused by multiple interference sources and node movement in more detail, and can provide more information for channel selection. However, existing channel interference prediction methods only predict single-step interference. If the current working channel is predicted to have interference, a channel is immediately selected for switching, without taking future interference into consideration, resulting in frequent channel switching, increasing the complexity and overhead of network management, and reducing communication efficiency. Summary of the invention

[0005] The purpose of the present invention is to provide an available link intelligent prediction and optimal selection method for unmanned cluster anti-interference communication. The method performs multi-channel multi-step interference power prediction through an LSTM-KAN network, combines the long-term memory advantage of LSTM and the ability of KAN to avoid catastrophic forgetting, significantly improves the accuracy of interference power prediction, and improves the mean square error performance by 25.8%. Subsequently, a classification sorting genetic algorithm (CSGA) is used to solve and construct a dual-objective optimization model that minimizes the number of channel switching times and average interference power. Compared with traditional optimization algorithms, better solution results can be obtained. Simulation results show that the method proposed in the present invention can significantly reduce the number of channel switching times, increase throughput, and reduce packet loss rate compared to the random channel selection strategy of single-step interference prediction, and has important application value and reference significance.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication, the method comprising the following steps:

[0008] Step 1) obtaining a multi-channel interference power matrix;

[0009] Step 2) establishing an LSTM-KAN interference prediction model, wherein the LSTM-KAN interference prediction model takes a multi-channel interference power matrix as a time series input and outputs a channel interference power prediction result;

[0010] Step 3) Use the LSTM-KAN interference prediction model to predict the channel interference power of multiple time slots in the future;

[0011] Step 4) modeling channel selection as a dual-objective optimization model of minimizing the number of channel switching times and minimizing the average interference power of the channel;

[0012] Step 5) Solve the dual-objective optimization model based on the classification and sorting genetic algorithm, increase the probability of generating a solution with a low number of channel switching times by improving the crossover mutation strategy and the elite retention strategy, and make the solution with a low number of channel switching times be retained as much as possible during the iteration process, and finally generate an optimized channel switching and interference power solution.

[0013] The LSTM-KAN interference prediction model includes a sequence input layer, an LSTM layer, a KAN layer and an output layer. The input multi-channel interference power matrix is ​​input into the LSTM layer to extract high-dimensional time series features. The high-dimensional time series features are reduced in dimension through the KAN layer, and the high-dimensional features are mapped to the target prediction dimension. Finally, the interference power prediction result is generated through the output layer.

[0014] The KAN layer is set to two layers, the high-dimensional time series features are input into the first KAN layer, the internal function features are activated and learned; and the output of the first KAN layer is transferred to the second KAN layer, the external function features are activated and learned.

[0015] In the step 3), a recursive prediction method is used to iterate the channel interference power prediction value of each time step as the input of the LSTM-KAN interference prediction model of the next time step until the prediction values ​​of all time slots are generated.

[0016] The recursive prediction process is expressed as:

[0017] P i =[p i-L+1 ,p i-L+2 ...p i ]

[0018]

[0019] Among them, p i =[p i1 ,p i2 ,...p iN ] is the N channel interference power sensed by the sensing node in time slot i, p in is the interference power of channel n in the i-th time slot, P i is the multi-channel interference power matrix of L time slots before time slot i, To utilize P i Predict the channel interference power matrix of the next T time slots, where f is the LSTM-KAN interference prediction model.

[0020] The objective function of the dual-objective optimization model is expressed as:

[0021]

[0022] Among them, X i The data channel number of the unmanned trunking operation in time slot i, X i =[X i+1 ,X i+2 ...X i+T ] is the data channel number of the unmanned cluster in the next T time slots; the data channel of the unmanned cluster in time slot i is channel c, that is, X i =c;I i Indicates whether the data channel of the unmanned trunking in time slot i is switched. When the channels in time slot i are different, I i =1, otherwise I i =0;S i is the number of channel switching times in the next T time slots starting from time slot i; PA i is the average channel interference power in the next T time slots.

[0023] The classification and sorting genetic algorithm improves the crossover mutation strategy during solution space evolution to increase the probability of generating low channel switching solutions, that is, crossover is performed at channel discontinuities to protect continuous genes, and during mutation, some discontinuous channels are randomly transformed into the same channels to more quickly generate solutions with fewer channel switching times; when the solution space is retained, a classification and sorting elite retention strategy is designed to retain the low channel switching solutions as much as possible, that is, classification is performed according to the number of channel switching times, and the solution with the smallest average interference power in each category is retained.

[0024] The classification and sorting genetic algorithm performs the following steps:

[0025] Step 51) Initialize the population and calculate the fitness: randomly generate M individuals that meet the constraints as the initial parent population, and calculate the fitness value of each individual in the population according to the objective function of the dual-objective optimization model;

[0026] Step 52) Select crossover mutation to generate a new population: select two parents through a binary selection method, perform crossover mutation operations on the selected parents to generate new offspring, and randomly adjust the channels in the offspring through mutation operations to increase the probability of the occurrence of solutions with a low number of switching times;

[0027] Step 53) Classification and sorting elite retention: adopt the classification and sorting elite retention strategy, classify the solutions in the population according to the number of channel switching times, and retain the solution with the lowest channel interference power in each category, and give priority to retaining the solution with low switching number;

[0028] Step 54) repeats step 52) to step 53) in a loop, and exits the loop when the number of iterations is greater than the maximum number of iterations constraint;

[0029] Step 55) Extracting the Pareto front: traverse the last generation population, retain the solution with the minimum average interference power under each channel switching number, then perform non-dominated sorting on the solution set consisting of the retained solutions, and extract the solution with a dominated number of 0 to form the Pareto front solution set;

[0030] Step 56) Select the final solution: If the number of solutions in the Pareto front solution set is less than or equal to 2, directly take the one with the least number of channel switching times as the final solution; if the number of solutions in the Pareto front solution set is greater than 2, sort the solutions in the Pareto front solution set according to the number of channel switching times, calculate the ratio of the difference between the average interference power of the channel of each solution and the previous solution and the difference in the number of channel switching times, that is, the slope value, and select the one with the largest absolute value of the slope as the final solution.

[0031] The step 52) comprises the following steps:

[0032] Step 521) using binary selection method to select two parents P1 and P2;

[0033] Step 522) crossover the two selected parents to find the non-continuous position set S of P1, and randomly select a position k in the set S to make P1 and P2 perform chromosome crossover at k to generate offspring C1 and C2;

[0034] Step 523) randomly select a position l in the set S, randomly generate an integer n, 1≤n≤Tl, randomly select a channel m from the N channels, and let P1 mutate to m at l~l+n to generate a new child C3, where T is the number of future time slots predicted by the LSTM-KAN interference prediction model;

[0035] Step 524) Repeat From step 521) to step 523), selection crossover mutation is performed, and the generated offspring and parent generations are merged into a new generation population.

[0036] The step 53) comprises the following steps:

[0037] Step 531) traverse the population, classify the solutions with the same number of channel switching times into one category, classify them by the number of channel switching times and count the number of solutions in each category;

[0038] Step 532) Evenly distribute the number of elites to be retained for each type of channel switching number of solutions according to the ratio λ of the population size to the type of channel switching number. If the number of solutions in a certain type is less than or equal to λ, the number of elites to be retained is the number of solutions in that type; if the number of solutions in a certain type is greater than λ, the number of elites to be retained is the ratio of the remaining population size to the remaining number of types;

[0039] Step 533) sorts the solution sets under each type of channel switching times according to the average channel interference power from small to large, and prioritizes the solutions with low average channel interference power according to the number of retained elites calculated in step 532) to form a new generation population.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] (1) The dual-target channel selection based on multi-step interference prediction in the present invention takes into account future interference conditions, and can minimize the number of channel switching while minimizing the average interference power of the channel, thereby avoiding the problem of frequent channel switching and reducing the complexity and overhead of network management;

[0042] (2) The LSTM-KAN network used in the present invention can more accurately predict multi-channel multi-step interference power and improve the accuracy and reliability of channel selection by integrating the advantages of long-term memory and avoiding catastrophic forgetting;

[0043] (3) The classification and sorting genetic algorithm (CSGA) adopted by the present invention improves the crossover mutation strategy and the elite retention strategy. Compared with the existing algorithm, it can efficiently converge to the Pareto frontier and efficiently solve the low channel switching solution;

[0044] (4) Compared with traditional methods, the present invention can effectively reduce the number of channel switching by 26%, increase the throughput by 8%, and significantly reduce the packet loss rate by 95%, thereby improving the communication performance of unmanned clusters in an anti-interference environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0046] Figure 2 This is a performance comparison chart of the LSTM-KAN network in the present invention and other networks;

[0047] Figure 3 This is a performance comparison chart of the CSGA algorithm in the present invention and other algorithms;

[0048] Figure 4 This is a performance comparison chart between the method of the present invention and the traditional algorithm. DETAILED DESCRIPTION

[0049] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0050] This embodiment provides a method for intelligent prediction and optimal selection of available links for anti-interference communication in unmanned clusters, aiming to solve the problem that existing channel interference prediction methods do not take future interference situations into consideration, resulting in frequent channel switching, increased network management complexity, and reduced communication efficiency. This method is based on the multi-step interference situation prediction and multi-objective optimization link selection architecture of the LSTM-KAN network. When the drone cluster faces electromagnetic interference challenges, interference suppression and communication stability are achieved through intelligent prediction and optimized link selection. This method uses the LSTM-KAN network to predict multi-channel multi-step interference power, and achieves optimal link selection through a dual-objective optimization model, further improving the performance of the communication system. Figure 1 As shown, the method includes:

[0051] Interference situation prediction stage: The LSTM-KAN network is used to perform multi-step predictions on multi-channel interference power to capture the trend of channel interference changes. The LSTM-KAN network combines the long-term memory advantage of LSTM with the ability of KAN to avoid catastrophic forgetting, thereby more accurately predicting future interference power. By comparison, the prediction accuracy of the LSTM-KAN network has been significantly improved compared with traditional LSTM, KAN and MLP networks, and the mean square error has been reduced by 25.8%.

[0052] Link selection optimization stage: Based on the interference power prediction results, a dual-objective optimization model is constructed to minimize the number of channel switching times and minimize the average channel interference power, and the number of channel switching times and the average channel interference power are optimized; then, the classification sorting genetic algorithm (CSGA) is used to solve the model for link selection, and the genetic algorithm is used to optimize the link configuration under the interference environment, and finally the optimal link selection scheme is obtained to maximize the communication performance, which can obtain a better solution effect than traditional algorithms (such as NSGA2, SMPSO and MOEA / D, etc.). This method not only effectively reduces the number of channel switching times, but also ensures that the channel selection is more efficient.

[0053] The application scenario of the method of the present invention is the scenario in which an unmanned cluster performs link switching selection when facing strong electromagnetic interference. Unmanned cluster nodes have independent control channels and data channels. The control channel rate is relatively low, and anti-interference can be achieved through large-bandwidth spread spectrum and frequency hopping; the data channel rate is high, and the broadband frequency hopping anti-interference gain is low, and channel switching is required to achieve anti-interference. The interference equipment is far away, and the interference situation perceived by the dense unmanned cluster is similar. The interference situation of multiple channels can be perceived in parallel through division of labor and cooperation. The perception node reports the perception information to the perception center node through the control channel for interference prediction and channel selection, and sends the channel selection information to the perception node through the control channel to command it to switch the data channel to avoid interference.

[0054] Specifically, the method comprises the following steps:

[0055] Step 1) Obtain a multi-channel interference power matrix.

[0056] Step 2) Establish an LSTM-KAN interference prediction model. The LSTM-KAN interference prediction model takes the multi-channel interference power matrix as the time series input and outputs the channel interference power prediction result.

[0057] The LSTM-KAN interference prediction model includes a sequence input layer, an LSTM layer, a KAN layer and an output layer. The input multi-channel interference power matrix is ​​input into the LSTM layer to extract the long-term information in the sequence, and the high-dimensional time series features are output. The high-dimensional time series features are reduced in dimension through the KAN layer, and the high-dimensional features are mapped to the target prediction dimension. Finally, the interference power prediction result is generated through the output layer.

[0058] In this embodiment, the KAN layer is set to two layers, and the high-dimensional time series features are input into the first KAN layer to activate and learn the internal function features; then the output of the first KAN layer is passed to the second KAN layer to activate and learn the external function features.

[0059] Step 3) Use the LSTM-KAN interference prediction model to predict the channel interference power of multiple time slots in the future.

[0060] In this embodiment, a recursive prediction method is used to iterate the channel interference power prediction value of each time step as the input of the LSTM-KAN interference prediction model of the next time step until the prediction values ​​of all time slots are generated. Specifically, the prediction value of the second step is first generated using the prediction result of the first step, and then the prediction results of the subsequent time slots are generated through a recursive process, and finally the power matrix of the multi-step prediction is obtained. The process is expressed as:

[0061] P i =[p i-L+1 ,p i-L+2 ...p i ]

[0062]

[0063] Among them, p i =[p i1 ,p i2 ,...p iN ] is the N channel interference power sensed by the sensing node in time slot i, p in is the interference power of channel n in the i-th time slot, P i is the multi-channel interference power matrix of L time slots before time slot i, To utilize P i Predict the channel interference power matrix of the next T time slots, where f is the LSTM-KAN interference prediction model.

[0064] Step 4) Model the channel selection as a dual-objective optimization model of minimizing the number of channel switching times and minimizing the average channel interference power.

[0065] Excessive interference power will reduce network throughput, and frequent channel switching will increase network management overhead. Therefore, these two factors should be balanced to select channels. The objective function of the dual-objective optimization model is expressed as:

[0066]

[0067] Among them, X i The data channel number of the unmanned trunking operation in time slot i, X i =[X i+1 ,X i+2 ...Xi+T ] is the data channel number of the unmanned cluster in the next T time slots; the data channel of the unmanned cluster in time slot i is channel c, that is, X i =c;I i Indicates whether the data channel of the unmanned trunking in time slot i is switched. When the channels in time slot i are different, I i =1, otherwise I i =0;S i is the number of channel switching times in the next T time slots starting from time slot i; PA i is the average channel interference power in the next T time slots.

[0068] Step 5) Solve the dual-objective optimization model based on the classification and sorting genetic algorithm, increase the probability of generating a solution with a low number of channel switching times by improving the crossover mutation strategy and the elite retention strategy, and make the solution with a low number of channel switching times be retained as much as possible during the iteration process, and finally generate an optimized channel switching and interference power solution.

[0069] In the channel selection problem, since the number of channel switching is required to be small, the solution vector should be kept continuous as much as possible. The random evolution strategy of the traditional algorithm may lead to a low probability of low channel switching solutions, and these solutions may be eliminated due to the effect of the elite retention strategy, so it is difficult to solve a better solution. The proposed classification and sorting genetic algorithm improves the crossover mutation strategy and the elite retention strategy, and improves the performance of solving low channel switching solutions. Specifically, when the solution space evolves, the crossover mutation strategy is improved to increase the generation probability of low channel switching solutions, that is, crossover at the discontinuous part of the channel to protect the continuous gene, and when mutating, some discontinuous channels are randomly transformed into the same channel to generate solutions with fewer channel switching times faster; when the solution space is retained, the elite retention strategy of classification and sorting is designed to make the low channel switching solutions as much as possible retained, that is, classified according to the number of channel switching, and retain the solution with the smallest average interference power in each category. The classification retention method gives the solution with fewer channel switching times an opportunity to evolve, so that the channel average interference power target can also evolve and converge to a smaller value, improving the performance of the channel selection process. The specific steps of the classification and sorting genetic algorithm are as follows:

[0070] Step 51) Initialize the population and calculate the fitness: randomly generate M individuals that meet the constraints as the initial parent population, and calculate the fitness value of each individual in the population according to the objective function of the dual-objective optimization model;

[0071] Step 52) Select crossover mutation to generate a new population: select two parents through a binary selection method, perform crossover mutation operations on the selected parents to generate new offspring, and randomly adjust the channels in the offspring through mutation operations to increase the probability of the occurrence of solutions with a low number of switching times;

[0072] Step 53) Classification and sorting elite retention: adopt the classification and sorting elite retention strategy, classify the solutions in the population according to the number of channel switching times, and retain the solution with the lowest channel interference power in each category, and give priority to retaining the solution with low switching number;

[0073] Step 54) repeats step 52) to step 53) in a loop, and exits the loop when the number of iterations is greater than the maximum number of iterations constraint;

[0074] Step 55) Extracting the Pareto front: traverse the last generation population, retain the solution with the minimum average interference power under each channel switching number, then perform non-dominated sorting on the solution set consisting of the retained solutions, and extract the solution with a dominated number of 0 to form the Pareto front solution set;

[0075] Step 56) Select the final solution: If the number of solutions in the Pareto front solution set is less than or equal to 2, directly take the one with the least number of channel switching times as the final solution; if the number of solutions in the Pareto front solution set is greater than 2, sort the solutions in the Pareto front solution set according to the number of channel switching times, calculate the ratio of the difference between the average interference power of the channel of each solution and the previous solution and the difference in the number of channel switching times, that is, the slope value, and select the one with the largest absolute value of the slope as the final solution.

[0076] In one embodiment, step 52) includes the following steps:

[0077] Step 521) using binary selection method to select two parents P1 and P2;

[0078] Step 522) crossover the two selected parents to find the non-continuous position set S of P1, and randomly select a position k in the set S to make P1 and P2 perform chromosome crossover at k to generate offspring C1 and C2;

[0079] Step 523) randomly select a position l in the set S, randomly generate an integer n, 1≤n≤Tl, randomly select a channel m from the N channels, and let P1 mutate to m at l~l+n to generate a new child C3, where T is the number of future time slots predicted by the LSTM-KAN interference prediction model;

[0080] Step 524) Repeat From step 521) to step 523), selection crossover mutation is performed, and the generated offspring and parent generations are merged into a new generation population.

[0081] In one embodiment, step 53) includes the following steps:

[0082] Step 531) traverse the population, classify the solutions with the same number of channel switching times into one category, classify them by the number of channel switching times and count the number of solutions in each category;

[0083] Step 532) Evenly distribute the number of elites to be retained for each type of channel switching number of solutions according to the ratio λ of the population size to the type of channel switching number. If the number of solutions in a certain type is less than or equal to λ, the number of elites to be retained is the number of solutions in that type; if the number of solutions in a certain type is greater than λ, the number of elites to be retained is the ratio of the remaining population size to the remaining number of types;

[0084] Step 533) sorts the solution sets under each type of channel switching times according to the average channel interference power from small to large, and prioritizes the solutions with low average channel interference power according to the number of retained elites calculated in step 532) to form a new generation population.

[0085] Based on the above method, this embodiment provides the following examples:

[0086] In the first step, each sensing node in the drone cluster monitors the surrounding electromagnetic environment through the control channel, and the sensing center obtains the interference intensity on 9 channels every 0.1 seconds;

[0087] The second step is to establish and train the LSTM-KAN interference prediction model;

[0088] In the third step, the perception center uses a recursive prediction method based on the trained LSTM-KAN interference prediction model and the multi-channel interference power information perceived in the current scenario to predict the next 30 multi-channel interferences using the past 30 multi-channel interference power data, and obtains a 30*9-dimensional prediction matrix

[0089] The fourth step is to use the power prediction matrix A dual-objective optimization model is established to minimize the number of channel switching times and minimize the average channel interference power.

[0090] The fifth step is to solve the dual-objective optimization model using the classification and sorting genetic algorithm. The specific steps are as follows:

[0091] a. Initialize the population and calculate the fitness: According to the optimization model, randomly generate 200 individuals that meet the constraints as the initial parent population, and calculate the fitness value of each individual in the population according to the model objective function.

[0092] b. Select crossover mutation to generate a new population. Repeat the method in step 52) above for 100 times, and finally merge the generated offspring and parent generations into a new generation population.

[0093] c. Perform classification and sorting according to the method in step 53) above and retain the elite.

[0094] d. Repeat steps b to c, and exit the loop when the number of iterations is greater than the maximum number of iterations 500.

[0095] e. Extract the Pareto front according to the method in step 55) above.

[0096] f. Select the final solution according to the method in step 56) above.

[0097] Figure 2 This is a performance comparison chart of the LSTM-KAN network in the present invention and other networks. Figure 3 This is a performance comparison chart of the CSGA algorithm in the present invention and other algorithms. Figure 4 This is a performance comparison chart between the method of the present invention and the traditional algorithm. The present invention proposes for the first time a channel selection anti-interference method based on multi-step interference prediction to minimize the number of channel switching times and minimize the average interference power of the channel, and designs an LSTM-KAN network for interference prediction and a CSGA algorithm to solve the dual-objective channel selection problem. The examples of the present invention show that compared with the traditional method, this method can reduce the number of channel switching times by 26%, increase the throughput by 8%, and reduce the packet loss rate by 95%, which proves the effectiveness of the anti-interference link selection method of the present invention and has important practical significance and application value.

[0098] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication, characterized in that: The method comprises the following steps: Step 1) obtaining a multi-channel interference power matrix; Step 2) establishing an LSTM-KAN interference prediction model, wherein the LSTM-KAN interference prediction model takes a multi-channel interference power matrix as a time series input and outputs a channel interference power prediction result; Step 3) Use the LSTM-KAN interference prediction model to predict the channel interference power of multiple time slots in the future; Step 4) modeling channel selection as a dual-objective optimization model of minimizing the number of channel switching times and minimizing the average interference power of the channel; Step 5) Solve the dual-objective optimization model based on the classification and sorting genetic algorithm, increase the probability of generating a solution with a low number of channel switching times by improving the crossover mutation strategy and the elite retention strategy, and make the solution with a low number of channel switching times be retained as much as possible during the iteration process, and finally generate an optimized channel switching and interference power solution.

2. According to claim 1, a method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication, characterized in that: The LSTM-KAN interference prediction model includes a sequence input layer, an LSTM layer, a KAN layer and an output layer. The input multi-channel interference power matrix is ​​input into the LSTM layer to extract high-dimensional time series features. The high-dimensional time series features are reduced in dimension through the KAN layer, and the high-dimensional features are mapped to the target prediction dimension. Finally, the interference power prediction result is generated through the output layer.

3. The method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication according to claim 2 is characterized in that: The KAN layer is set to two layers, and the high-dimensional time series features are input into the first KAN layer to activate and learn the internal function features; The output of the first KAN layer is then passed to the second KAN layer to activate and learn the external function features.

4. According to claim 1, a method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication, characterized in that: In the step 3), a recursive prediction method is used to iterate the channel interference power prediction value of each time step as the input of the LSTM-KAN interference prediction model of the next time step until the prediction values ​​of all time slots are generated.

5. The method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication according to claim 4 is characterized in that: The recursive prediction process is expressed as: P i =[p i-L+1 ,p i-L+2 ...p i ] Among them, p i =[p i1 ,p i2 ,...p iN ] is the N channel interference power sensed by the sensing node in time slot i, p in is the interference power of channel n in the i-th time slot, P i is the multi-channel interference power matrix of L time slots before time slot i, To utilize P i Predict the channel interference power matrix of the next T time slots, where f is the LSTM-KAN interference prediction model.

6. The method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication according to claim 5, characterized in that: The objective function of the dual-objective optimization model is expressed as: Among them, X i The data channel number of the unmanned trunking operation in time slot i, X i =[X i+1 ,X i+2 ...X i+T ] is the data channel number of the unmanned cluster in the next T time slots; the data channel of the unmanned cluster in time slot i is channel c, that is, X i =c;I i Indicates whether the data channel of the unmanned trunking in time slot i is switched. When the channels in time slot i are different, I i =1, otherwise I i =0;S i is the number of channel switching times in the next T time slots starting from time slot i; PA i is the average channel interference power in the next T time slots.

7. The method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication according to claim 1 is characterized in that: The classification and sorting genetic algorithm improves the crossover mutation strategy during solution space evolution to increase the probability of generating low channel switching solutions, that is, crossover is performed at channel discontinuities to protect continuous genes, and during mutation, some discontinuous channels are randomly transformed into the same channels to more quickly generate solutions with fewer channel switching times; when the solution space is retained, a classification and sorting elite retention strategy is designed to retain the low channel switching solutions as much as possible, that is, classification is performed according to the number of channel switching times, and the solution with the smallest average interference power in each category is retained.

8. The method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication according to claim 1 is characterized in that: The classification and sorting genetic algorithm performs the following steps: Step 51) Initialize the population and calculate the fitness: randomly generate M individuals that meet the constraints as the initial parent population, and calculate the fitness value of each individual in the population according to the objective function of the dual-objective optimization model; Step 52) Select crossover mutation to generate a new population: select two parents through a binary selection method, perform crossover mutation operations on the selected parents to generate new offspring, and randomly adjust the channels in the offspring through mutation operations to increase the probability of the occurrence of solutions with a low number of switching times; Step 53) Classification and sorting elite retention: adopt the classification and sorting elite retention strategy, classify the solutions in the population according to the number of channel switching times, and retain the solution with the lowest channel interference power in each category, and give priority to retaining the solution with low switching number; Step 54) repeats step 52) to step 53) in a loop, and exits the loop when the number of iterations is greater than the maximum number of iterations constraint; Step 55) Extracting the Pareto front: traverse the last generation population, retain the solution with the minimum average interference power under each channel switching number, then perform non-dominated sorting on the solution set consisting of the retained solutions, and extract the solution with a dominated number of 0 to form the Pareto front solution set; Step 56) Select the final solution: If the number of solutions in the Pareto front solution set is less than or equal to 2, directly select the one with the least number of channel switching times as the final solution; If the number of solutions in the Pareto front solution set is greater than 2, sort the solutions in the Pareto front solution set according to the number of channel switching times, calculate the ratio of the difference between each solution and the previous solution's channel average interference power and the difference in the number of channel switching times, that is, the slope value, and select the one with the largest absolute value of the slope as the final solution.

9. The method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication according to claim 8, characterized in that: The step 52) comprises the following steps: Step 521) using binary selection method to select two parents P1 and P2; Step 522) crossover the two selected parents to find the non-continuous position set S of P1, and randomly select a position k in the set S to make P1 and P2 perform chromosome crossover at k to generate offspring C1 and C2; Step 523) randomly select a position l in the set S, randomly generate an integer n, 1≤n≤Tl, randomly select a channel m from the N channels, and let P1 mutate to m at l~l+n to generate a new child C3, where T is the number of future time slots predicted by the LSTM-KAN interference prediction model; Step 524) Repeat From step 521) to step 523), selection crossover mutation is performed, and the generated offspring and parent generations are merged into a new generation population.

10. The method for intelligent prediction and optimal selection of available links for unmanned cluster anti-interference communication according to claim 8, characterized in that: The step 53) comprises the following steps: Step 531) traverse the population, classify the solutions with the same number of channel switching times into one category, classify them by the number of channel switching times and count the number of solutions in each category; Step 532) Evenly distribute the number of elites to be retained for each type of channel switching number of solutions according to the ratio λ of the population size to the type of channel switching number. If the number of solutions in a certain type is less than or equal to λ, the number of elites to be retained is the number of solutions in that type; if the number of solutions in a certain type is greater than λ, the number of elites to be retained is the ratio of the remaining population size to the remaining number of types; Step 533) sorts the solution sets under each type of channel switching times according to the average channel interference power from small to large, and prioritizes the solutions with low average channel interference power according to the number of retained elites calculated in step 532) to form a new generation population.

Images