Method for intelligent prediction and optimal selection of available links for anti-jamming communication of unmanned cluster

Abstract

The present application relates to a kind of unmanned cluster interference communication-oriented available link intelligent prediction and optimal selection method, comprising the following steps: obtaining multi-channel interference power matrix;Establishing LSTM-KAN interference prediction model;The channel interference power of future multiple time slots is predicted using LSTM-KAN interference prediction model;Channel selection is modeled as the double-objective optimization model of minimizing channel switching number and minimizing channel average interference power;Based on classification sorting genetic algorithm, the double-objective optimization model is solved, the generation probability of low channel switching number solution is increased by improving cross variation strategy and elite reservation strategy, and the low channel switching number solution is as far as possible reserved in iteration process, and finally the optimized channel switching and interference power solution are generated.Compared with prior art, the present application has the advantages of significantly reducing channel switching number, improving throughput and reducing packet loss rate, and has important application value for unmanned cluster communication in strong electromagnetic interference environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of anti-jamming communication of unmanned cluster, and particularly to a method for intelligent prediction and optimal selection of available link for anti-jamming communication of unmanned cluster. BACKGROUND

[0002] In recent years, unmanned clusters are widely used in reconnaissance, attack, rescue and other types of tasks. The communication network connecting multiple platforms is the basis for the unmanned cluster to perform tasks cooperatively. However, in the counter-environment, the unmanned cluster often faces strong electromagnetic interference, and requires the communication network to have the ability to resist strong interference. In addition to the commonly used frequency hopping, spread spectrum and directional antenna anti-interference, real-time sensing and predicting channel situation and optimally selecting working channels is an important anti-interference method.

[0003] According to different prediction contents, the interference channel prediction can be divided into two categories: channel occupation prediction and channel interference power prediction. Channel occupation prediction is a binary classification prediction of whether the channel is interfered or not. The prediction methods include regression analysis, hidden Markov model and neural network. 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 dependence characteristic leads to low prediction accuracy; the prediction method based on MLP and LSTM neural network has become an important technology for channel situation prediction, but it needs a sufficient data set for training. Channel interference power prediction is a regression prediction of the interference power value sensed by the frequency spectrum. The prediction method is mainly neural network method, such as using LSTM network to realize power spectrum density prediction and randomly selecting a channel with interference power below the threshold to access, using LSTM network to realize channel power prediction and selecting the channel with the maximum expected throughput to access.

[0004] Channel occupation prediction depends on the accuracy of spectrum sensing decision, and the binary classification characteristic leads to that the channel selection can only randomly select a channel without interference. 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, the existing channel interference prediction method only predicts single-step interference. If the current working channel is predicted to have interference, a channel will be selected for switching immediately, without considering the future interference, which leads to frequent channel switching, increases the complexity and overhead of network management, and reduces the communication efficiency. SUMMARY

[0005] The application aims to provide a method for intelligent prediction and optimal selection of available links for anti-interference communication of unmanned clusters.

[0006] The application can be achieved by the following technical solutions.

[0007] A method for intelligent prediction and optimal selection of available links for anti-interference communication of unmanned clusters comprises the following steps.

[0008] Step 1) Obtain a multi-channel interference power matrix.

[0009] Step 2) Establish an LSTM-KAN interference prediction model, which takes the multi-channel interference power matrix as time series input and outputs channel interference power prediction results.

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

[0011] Step 4) Model channel selection as a double-objective optimization model for minimizing channel switching times and minimizing average channel interference power.

[0012] Step 5) Solve the double-objective optimization model based on the classification and sorting genetic algorithm, improve the crossover and mutation strategies and the elite reservation strategy to increase the generation probability of low channel switching time solutions, and make the low channel switching time solutions be reserved as much as possible in the iteration process, and finally generate optimized channel switching and interference power solutions.

[0013] The LSTM-KAN interference prediction model comprises 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 sequence features, the high-dimensional time sequence features are mapped to the target prediction dimension through the KAN layer, and finally the interference power prediction results are generated through the output layer.

[0014] The KAN layer is arranged in two layers, the high-dimensional time sequence feature is input into the first layer KAN layer, the internal function feature is activated and learned, and the output of the first layer KAN layer is transmitted to the second layer KAN layer, and the external function feature is activated and learned.

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

[0016] The recursive prediction process is represented as:

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

[0018]

[0019] Wherein, p i =[p i1 ,p i2 ,...p iN ] is the N channel interference power obtained by the sensing node in the i time slot, p in is the interference power of channel n in the i time slot, P i is the multi-channel interference power matrix of the i time slot and the previous L time slots, is the channel interference power matrix of the future T time slots predicted by P i , and f is the LSTM-KAN interference prediction model.

[0020] The objective function of the double-target optimization model is represented as:

[0021]

[0022] Wherein, X i is the data channel number of the i time slot unmanned cluster, X i =[X i+1 ,X i+2 ...X i+T ] is the data channel number of the future T time slots of the unmanned cluster; the data channel of the i time slot unmanned cluster is channel c, that is, X i =c; I i represents whether the data channel of the i time slot unmanned cluster is switched, when the i time slot channel is different, I i =1, otherwise I i =0; S i is the channel switching number of the i time slot and the future T time slots; PA i is the average channel interference power of the future T time slots.

[0023] The classification and sorting genetic algorithm improves the crossover and mutation strategy when evolving the solution space to increase the probability of generating low channel switching solutions, i.e., crossing at the discontinuous channels to protect the continuous genes, and randomly transforming part of the discontinuous channels into the same channel when mutating to produce solutions with fewer channel switching times faster; the classification and sorting elitist reservation strategy is designed when preserving the solution space to make low channel switching solutions be preserved as much as possible, i.e., according to the channel switching times classification, the solution with the minimum average interference power in each class is reserved.

[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 satisfying the constraint conditions to become the initial parent population, and calculate the fitness value of each individual in the population according to the objective function of the double-objective optimization model;

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

[0027] Step 53) classification and sorting elitist reservation: adopt the classification and sorting elitist reservation strategy, classify the solutions in the population according to the channel switching times, and reserve the solution with the lowest channel interference power in each class, preferentially reserve the low switching number solution;

[0028] Step 54) repeat steps 52) to 53) in a loop, and exit the loop when the iteration number is greater than the maximum iteration number constraint;

[0029] Step 55) extract the Pareto front: traverse the last generation population, reserve the solution with the minimum average interference power under each channel switching times, and perform non-dominated sorting on the solution set composed of the reserved solutions to extract the solutions 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 solution with the least 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 channel switching times, calculate the ratio of the difference between the average channel interference power of each solution and the previous solution to the difference between the channel switching times, i.e., the slope value, and select the solution with the maximum absolute value of the slope as the final solution.

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

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

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

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

[0035] Step 524) repeat Sub-step 521) to sub-step 523) selection, crossing and mutation, and combining the generated offspring and the parents to form 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 a class, classify according to the number of channel switching times, and count the number of solutions in each class;

[0038] Step 532) evenly distribute the elite amount to be reserved for each type of channel switching time solution according to the ratio λ of the population size to the number of channel switching time categories, if the number of solutions in a certain class is less than or equal to λ, the elite amount to be reserved is the number of solutions in the class; if the number of solutions in a certain class is greater than λ, the elite amount to be reserved is the ratio of the remaining population size to the remaining number of categories;

[0039] Step 533) sort the solution set under each type of channel switching time according to the average channel interference power from small to large, and preferentially reserve the solutions with low channel average interference power to form a new generation population according to the reserved elite amount calculated in step 532).

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

[0041] (1) The double-target channel selection based on multi-step interference prediction in the present application considers future interference, can minimize the channel switching times while minimizing the average channel interference power, avoids the problem of frequent channel switching, and reduces the complexity and overhead of network management;

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

[0043] (3) The classification sorting genetic algorithm (CSGA) used in the application improves the crossover mutation strategy and the elite reservation strategy, and compared with existing algorithms, can efficiently converge to the Pareto front and efficiently solve low channel switching solutions;

[0044] (4) Compared with the traditional method, the application can effectively reduce the channel switching times by 26%, increase the throughput by 8%, and greatly reduce the packet loss rate by 95%, thereby improving the communication performance of the unmanned cluster in the anti-interference environment. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 is a method flowchart of the application;

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

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

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

[0049] The application will be described in detail below in conjunction with the drawings and specific embodiments. The present embodiment is implemented on the basis of the technical solution of the application, and gives a detailed implementation and specific operation process, but the protection scope of the application is not limited to the following examples.

[0050] The present embodiment provides a method for intelligent prediction and optimal selection of available links for anti-interference communication of unmanned clusters, which aims to solve the problem that the existing channel interference prediction method does not consider future interference, resulting in frequent channel switching, increasing network management complexity and reducing communication efficiency. The method is based on the multi-step interference situation prediction of the LSTM-KAN network and the multi-objective optimization link selection architecture. When the unmanned cluster is facing electromagnetic interference challenges, intelligent prediction and optimal link selection are used to realize interference suppression and communication stability. The method uses the LSTM-KAN network to predict the multi-channel multi-step interference power, and realizes optimal link selection through a double-objective optimization model, further improving the performance of the communication system. As shown in Figure 1 The method comprises the following steps:

[0051] Interference situation prediction stage: use LSTM-KAN network to multi-step predict multi-channel interference power, capture channel interference trend. LSTM-KAN network combines the long-term memory advantage of LSTM and the ability to avoid catastrophic forgetting of KAN, so as to more accurately predict future interference power. Through comparison, the prediction accuracy of LSTM-KAN network is significantly improved compared with traditional LSTM, KAN and MLP network, and the mean square error is reduced by 25.8%.

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

[0053] The application scenario of the method of the application is the scenario of link switching selection of unmanned cluster in the face of strong electromagnetic interference. The unmanned cluster nodes have independent control channels and data channels. The control channel has low rate and can realize anti-interference through wideband spread spectrum and frequency hopping; the data channel has high rate and low anti-interference gain of wideband frequency hopping, and needs to realize anti-interference through channel switching. The interference equipment is far away, and the dense unmanned cluster can perceive similar interference situation through division of labor and parallel perception of multiple channel interference. The perception node reports the perception information to the perception center node for interference prediction and channel selection through the control channel, and issues the channel selection information to the perception node through the control channel to command the perception node to switch the data channel to avoid interference.

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

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

[0056] Step 2) establishing an LSTM-KAN interference prediction model, the LSTM-KAN interference prediction model taking the multi-channel interference power matrix as a time sequence input and outputting channel interference power prediction results.

[0057] The LSTM-KAN interference prediction model comprises 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 long-term information in the sequence, and high-dimensional time sequence features are output, the high-dimensional time sequence features are mapped to a target prediction dimension through the KAN layer, and finally the interference power prediction results are generated through the output layer.

[0058] In this embodiment, the KAN layer is set to two layers, the high-dimensional time sequence feature is input into the first layer KAN layer, and the internal function feature is activated and learned; and then the output of the first layer KAN layer is transmitted to the second layer KAN layer, and the external function feature is activated and learned.

[0059] Step 3) predicting the channel interference power of future time slots by using the LSTM-KAN interference prediction model.

[0060] In this embodiment, the recursive prediction method is used, and the channel interference power prediction value of each time step is used as the input of the next time step LSTM-KAN interference prediction model for iteration until the prediction values of all time slots are generated. Specifically, the second step prediction value is generated by using the first step prediction result, and the prediction result of the subsequent time slot is generated through the recursive process, and finally the power matrix of multi-step prediction is obtained, and the process is represented as:

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

[0062]

[0063] wherein p i =[p i1 ,p i2 ,...p iN ] is the N channel interference power perceived by the perception node at the i time slot, p in is the interference power of channel n at the i time slot, P i is the multi-channel interference power matrix of the i time slot and the previous L time slots, is the channel interference power matrix of the future T time slots predicted by using P i , and f is the LSTM-KAN interference prediction model.

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

[0065] The excessive interference power reduces the network throughput, and the frequent channel switching increases the network management overhead, so the channel selection is balanced by considering these two factors, and the objective function of the dual-objective optimization model is represented as:

[0066]

[0067] wherein X i is the data channel number of the unmanned cluster working at the i time slot, X i =[X i+1 ,X i+2 ...Xi+T ] is the data channel number that the unmanned cluster works in the future T time slots; i is the data channel that the unmanned cluster works in the i time slot, i.e., X i = c; i i indicates whether the data channel that the unmanned cluster works in the i time slot is switched, i.e., when the channels in the i time slot are different, I i = 1, otherwise I i = 0; S i is the number of channel switching in the future T time slots from the i time slot; PA i is the average interference power of the channel in the future T time slots.

[0068] Step 5) solving the double-objective optimization model based on the classification and sorting genetic algorithm, improving the crossover and mutation strategies and the elite reservation strategy to increase the generation probability of the low channel switching number solution and make the low channel switching number solution be reserved as much as possible in the iteration process, and finally generating the optimized channel switching and interference power solution.

[0069] In the channel selection problem, since the channel switching number is required to be small, the solution vector should be kept continuous as much as possible, and the random evolution strategy of the traditional algorithm may lead to a low probability of the appearance of the low channel switching solution, and these solutions may be eliminated due to the effect of the elite reservation strategy, so it is difficult to solve the optimal solution. The proposed classification and sorting genetic algorithm improves the crossover and mutation strategies and the elite reservation strategy, and improves the solving performance of the low channel switching solution. Specifically, the crossover and mutation strategies are improved to increase the generation probability of the low channel switching solution when the solution space evolves, i.e., the non-continuous channels are crossed to protect the continuous genes, and part of the non-continuous channels are randomly transformed into the same channel to produce the solution with a smaller channel switching number faster when the solution space is reserved. The classification and sorting elite reservation strategy is designed to make the low channel switching solution be reserved as much as possible, i.e., according to the channel switching number classification, the solution with the minimum average interference power in each class is reserved, and the classification and reservation method gives the evolution opportunity to the solution with a smaller channel switching number, so that the channel average interference power target can also evolve to a smaller value, and the performance of the channel selection process is improved. The specific steps of the classification and sorting genetic algorithm are as follows:

[0070] Step 51) initializing the population and calculating the fitness: randomly generating M individuals that meet the constraint conditions to become the initial parent population, and calculating the fitness value of each individual in the population according to the objective function of the double-objective optimization model;

[0071] Step 52) selecting crossover and mutation to generate a new population: selecting two parents by binary selection method, performing crossover and mutation operation on the selected parents to generate new offspring, and randomly adjusting the channels in the offspring through the mutation operation to increase the appearance probability of the low switching number solution;

[0072] Step 53) Classification sorting elite reservation: adopt the classification sorting elite reservation strategy, classify the solutions in the population according to the channel switching times, and reserve the solution with the lowest channel interference power in each category, and preferentially reserve the low switching times solution;

[0073] Step 54) Repeat step 52) to step 53), and when the iteration number is greater than the maximum iteration number constraint, exit the loop;

[0074] Step 55) Extract the Pareto front: traverse the last generation population, reserve the solution with the smallest average interference power under each channel switching times, and then sort the reserved solution set, and extract the solution with the 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 solution with the least 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 channel switching times, calculate the ratio of the difference between the channel average interference power of each solution and the previous solution to the difference between the channel switching times, that is, the slope value, and select the solution with the largest absolute value of the slope as the final solution.

[0076] In an embodiment, step 52) comprises the following steps:

[0077] Step 521) Select two parent solutions P1 and P2 by using binary selection method;

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

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

[0080] Step 524) Repeat The selection, crossing and mutation of sub-steps 521) to 523) will generate offspring and parents to be combined into a new generation population.

[0081] In an embodiment, step 53) comprises the following steps:

[0082] Step 531) Traverse the population, classify the solutions with the same channel switching times into a category, and classify and count the number of solutions in each category according to the channel switching times;

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

[0084] Step 533) Sort the solution set under each type of channel switching number according to the average channel interference power from small to large, and preferentially retain solutions with low channel average interference power to form a new generation population according to the number of elite solutions calculated in step 532).

[0085] Based on the above method, the following examples are provided in this embodiment:

[0086] First step, each sensing node in the UAV 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] Second step, establish and train the LSTM-KAN interference prediction model;

[0088] Third step, the sensing center uses the recursive prediction method to predict the future 30 multi-channel interference using the past 30 multi-channel interference power data based on the trained LSTM-KAN interference prediction model and the multi-channel interference power information sensed in the current scenario, and obtains a 30*9-dimensional prediction matrix

[0089] Fourth step, according to the power prediction matrix A dual-objective optimization model is established to minimize the number of channel switching and the average channel interference power.

[0090] Fifth step, use the classification sorting genetic algorithm to solve the dual-objective optimization model, the specific steps are as follows:

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

[0092] b. Selection, crossover and mutation to produce a new population. According to the method in step 52) above, repeat the selection, crossover and mutation 100 times, and finally combine the offspring and parents to form a new generation population.

[0093] c. Classify and sort the elite according to the method in step 53) above.

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

[0095] e. Extract the Pareto frontier as in step 55 above.

[0096] f. Select the final solution as in step 56 above.

[0097] Figure 2 For the performance comparison chart of LSTM-KAN network and other networks in the application, Figure 3 For the performance comparison chart of CSGA algorithm and other algorithms in the application, Figure 4 For the performance comparison chart of the method of the application and the traditional algorithm. The application first proposes a channel selection anti-jamming method based on multi-step interference prediction to minimize the number of channel switches and minimize the average channel interference power, 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 application show that compared with the traditional method, the method can reduce the number of channel switches by 26%, increase the throughput by 8%, and reduce the packet loss rate by 95%, proving the effectiveness of the anti-jamming link selection method of the application, which has important practical significance and application value.

[0098] The preferred embodiments of the application are described in detail above. It should be understood that those skilled in the art can make many modifications and changes without creative labor based on the concept of the application. Therefore, any technical solution obtained by logical analysis, reasoning, or limited experiment based on the existing technology according to the concept of the application shall be within the protection scope determined by the claims.

Claims

1. A method for intelligent prediction and optimal selection of available links for anti-jamming communication of unmanned clusters, 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 the multi-channel interference power matrix as a time sequence input and outputs a channel interference power prediction result; Step 3) predicting channel interference power of a plurality of future time slots by using the LSTM-KAN interference prediction model; Step 4) modeling channel selection as a double-objective optimization model for minimizing channel switching times and minimizing average channel interference power; Step 5) solving the double-objective optimization model based on a classification and sorting genetic algorithm, improving a crossover and mutation strategy and an elite reservation strategy to increase a generation probability of a low channel switching time solution and to make the low channel switching time solution be reserved as much as possible in an iteration process, and finally generating an optimized channel switching and interference power solution. 2.The method of claim 1, wherein, The LSTM-KAN interference prediction model comprises 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 sequence features, the high-dimensional time sequence features are mapped to a target prediction dimension through the KAN layer, and finally an interference power prediction result is generated through the output layer.

3. The method of claim 2, wherein, The KAN layer is provided as two layers, the high-dimensional time sequence features are input into the first layer KAN layer, internal function features are activated and learned; and the output of the first layer KAN layer is transmitted to the second layer KAN layer, external function features are activated and learned.

4. The method of claim 1, wherein, In step 3), a recursive prediction method is adopted, a channel interference power prediction value of each time step is taken as an input of a next time step LSTM-KAN interference prediction model for iteration until prediction values of all time slots are generated.

5. The method of claim 4, wherein, The recursive prediction process is represented as: P i = [p i-L+1 , p i-L+2 ... p i ] where p i = [p i1 ,p i2 ,...p iN ] is the N channel interference power perceived by the sensing node at the i time slot, p in is the interference power of channel n at the i time slot, P i is the multi-channel interference power matrix of the i time slot and the previous L time slots, is the channel interference power matrix of the future T time slots predicted by P i , and f is the LSTM-KAN interference prediction model.

6. The method of claim 5, wherein, The objective function of the double-objective optimization model is represented as: Wherein, X i is the data channel number of the i time slot unmanned cluster work, X i = [X i+1 , X i+2 ... X i+T ] is the data channel number of the unmanned cluster work in the future T time slots; the data channel of the i time slot unmanned cluster work is channel c, that is, X i =c; I i indicates whether the data channel of the i time slot unmanned cluster work is switched, when the i time slot channel is different, I i =1, otherwise I i =0; S i is the channel switching number of the i time slot to the future T time slots; PA i is the average interference power of the channel of the future T time slots.

7. The method of claim 1, wherein, The classification and sorting genetic algorithm improves a crossover and mutation strategy to increase a generation probability of a low channel switching solution when evolving in a solution space, that is, a non-continuous channel is crossed to protect a continuous gene, and a part of non-continuous channels is randomly transformed into the same channel to produce a solution with less channel switching times faster when mutating; and an elite reservation strategy of classification and sorting is designed to make a low channel switching solution be reserved as much as possible when reserving in a solution space, that is, according to channel switching times classification, a solution with minimum average interference power in each class is reserved.

8. The method of claim 1, wherein, The classification and sorting genetic algorithm performs the following steps: Step 51) initializing a population and calculating fitness: M individuals meeting constraint conditions are randomly generated as an initial parent population, and fitness values of each individual in the population are calculated according to the objective function of the double-objective optimization model; Step 52) selecting and mutating to generate a new population: two parents are selected by a binary selection method, a new offspring is generated by performing a crossover and mutation operation on the selected parents, and a channel in the offspring is randomly adjusted by a mutation operation to increase a generation probability of a low switching time solution; Step 53) classification and sorting elite reservation: an elite reservation strategy of classification and sorting is adopted, solutions in the population are classified according to channel switching times, and a solution with minimum channel interference power in each class is reserved, and a low switching time solution is preferentially reserved; Step 54) Cycle repeats step 52) to step 53), and the cycle is exited when the iteration number is greater than the maximum iteration number constraint; Step 55) Extracting the Pareto front: traversing the last generation population, retaining the solution with the minimum average interference power for each channel switching number, and then performing non-dominated sorting on the solution set composed of the retained solutions to extract the solutions with a dominated number of 0 to form the Pareto front solution set; Step 56) Selecting the final solution: if the number of solutions in the Pareto front solution set is less than or equal to 2, directly taking the solution with the minimum channel switching number as the final solution; If the number of solutions in the Pareto front solution set is greater than 2, sorting the solutions in the Pareto front solution set according to the channel switching number, calculating the ratio of the difference between the average channel interference power of each solution and the previous solution to the difference between the channel switching numbers, i.e. the slope value, and selecting the solution with the maximum absolute value of the slope as the final solution.

9. The method of claim 8, wherein, The step 52) comprises the following steps: Step 521) Selecting two parent generations P1 and P2 by binary selection method; Step 522) Crossing the selected two parent generations to find the non-continuous position set S of P1, and randomly selecting a position k in the set S to perform chromosome crossing at k to generate offspring C1 and C2; Step 523) Randomly selecting a position l in the set S, randomly generating an integer n, 1≤n≤T-l, and randomly selecting a channel m in the N channels, and letting P1 mutate to m at l~l+n to generate a new offspring C3, wherein T is the number of future time slots predicted by the LSTM-KAN interference prediction model; Step 524) Repeat The selection crossover of sub-steps 521) to 523) produces offspring and parents combined into a new generation population.

10. The method of claim 8, wherein, The step 53) comprises the following steps: Step 531) Traversing the population, classifying the solutions with the same channel switching number into one class, classifying according to the channel switching number and counting the number of solutions in each class; Step 532) According to the ratio λ of the population size to the number of channel switching number categories, evenly distribute the elite amount to be retained for each channel switching number solution, if the number of solutions in a certain class is less than or equal to λ, the elite amount to be retained is the number of solutions in that class; if the number of solutions in a certain class is greater than λ, the elite amount to be retained is the ratio of the remaining population size to the remaining number of categories; Step 533) Sorting the solution set under each channel switching number according to the channel average interference power from small to large, and retaining the solutions with low channel average interference power to form a new generation population according to the retained elite amount calculated in step 532).

Images