Large-model-driven multi-population mixed integer optimization data collection array layout method
Through the multi-group hybrid integer optimization method driven by large-model, the problems of insufficient operation coverage and large data transmission delay in the multi-drone data collection array layout are solved. The drone array is improved by using population cooperation and large-model knowledge migration. Energy efficiency and coverage enable more efficient data collection.
Patent Information
- Application Number
- CN202410997645.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-26
AI Technical Summary
There are problems of insufficient operation coverage and large data transmission delay in the multi-UAV data collection array layout, mainly due to the complexity of mixed integer optimization problems and the inefficiency of existing algorithms when dealing with constraints and integer variables.
Using a large-model-driven mixed integer optimization method, the two populations are randomly initialized and independently evolved, and the drone array layout is optimized using big-model knowledge migration and prediction, combined with championship selection, differential evolution algorithm and reflection repair strategy, dynamic updates are updated. and optimize populations to improve the quality and diversity of solutions.
It improves the operation coverage and data transmission efficiency of the drone data collection array, reduces average energy consumption, enhances the population's evolution ability in environmental changes, and forms a more efficient optimization effect.
Smart Images

Figure CN119026457A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of intelligent computing methods, and in particular relates to a data collection array layout method for large model driven multi-population mixed integer optimization. Background Art
[0002] In recent years, drones have become a data collection tool with great potential in IoT data collection systems. However, in order to efficiently utilize drones, the layout of drones needs to be reasonably optimized during the data collection process.
[0003] Mixed-Integer Optimization Problem (MIOP) is a type of optimization problem whose decision variables include both integer variables and continuous variables. This problem is widely present in practical applications. The challenge of mixed-integer optimization is that it is necessary to simultaneously deal with the combinatorial optimization of integer variables and the numerical optimization of continuous variables, which makes the problem more complex and diverse. As a key link in mixed-integer optimization problems, the efficiency of variable processing methods directly determines the performance of the algorithm. When dealing with mixed-integer programming problems, existing evolutionary algorithms usually combine evolutionary algorithms with constraint processing techniques and integer processing methods to deal with the constraints and integer restrictions contained in the problem. In terms of constraint processing techniques, typical methods include penalty function methods, feasible solutions are better than infeasible solutions methods, and multi-objective-based methods. In terms of integer processing methods, the most widely used methods are rounding and truncation methods, that is, directly using rounding operations to convert continuous variables into integer variables. In addition, there are some methods that directly design encoding and evolution operators in the integer space when dealing with integer variables, so that the algorithm can always complete evolution directly in the integer space.
[0004] Multi-UAV-assisted data collection array layout optimization is a complex optimization problem that involves the number, location, and array layout of UAVs, the number and location of IoT devices, and the allocation of data collection tasks. This type of problem has obvious mixed variable characteristics because it requires the simultaneous optimization of integer variables (such as the task allocation of UAVs) and continuous variables (such as the specific coordinates of the UAV array). Therefore, the average energy consumption is too high during array layout, resulting in low operation coverage and excessive data transmission delay. Summary of the invention
[0005] The purpose of the present invention is to provide a data collection array layout method driven by a large model and multiple populations with mixed integer optimization, which solves the problem of insufficient coverage of multi-UAV data collection array layout operations in the prior art.
[0006] The technical solution adopted by the present invention is a data collection array layout method for large model driven multi-population mixed integer optimization, which is specifically implemented according to the following steps:
[0007] Step 1: Model the multi-UAV-assisted IoT data collection system, abstract the system layout optimization model, and determine the optimization objectives and objective functions;
[0008] Step 2, randomly initialize two populations;
[0009] Step 3, evolve the two populations independently. Before population 2 converges to a promising feasible region, record the solutions with good target values but infeasible that are eliminated during the evolution process as surviving solutions.
[0010] Step 4, reinitialize population 2 using the surviving solution and the converged solution of population 2 as the preferred gene;
[0011] Step 5: The large model drives knowledge transfer, learning and predicting the individuals in the two populations that can provide better target values, and integrating the continuous decision variables of the high-quality individuals in population 2 into the high-quality individuals in population 1 to guide evolution;
[0012] Step 6, iterative evolution, dynamic update and optimization of the large model, and output of the population 1 converged solution.
[0013] The present invention is also characterized in that:
[0014] Step 1 is implemented according to the following steps:
[0015] Step 1.1: Abstract the distance calculation method between the data source device and the drone
[0016] Using the variable u j ∈{0, 1}(j=1, ..., N) indicates whether the jth UAV is used to collect data. If the jth UAV is used to collect data, then u j =1; otherwise u j =0; for each data source device, fix its coordinates at (x i ,y i , 0)(i=1,...,γ), the height of the drone from the ground is defined as H, and the coordinates of the jth drone are (X j , Y j ,H), the distance calculation method from the i-th IoT device to the j-th drone is:
[0017]
[0018] Step 1.2: Abstract the relationship model between the data source device and the drone
[0019] The relationship between the i-th IoT device and the j-th drone is defined as a ij , if a ij =1, it means that the i-th data source device sends data to the j-th UAV; otherwise, a ij =0; In the current array, each data source device selects the nearest drone to send data, so the relationship between the device and the drone is as follows:
[0020]
[0021] Each data source device sends data to only one drone, so the relationship between the device and the drone is as follows:
[0022]
[0023] Due to the limitation of data transmission bandwidth, each drone has at most N data receiving channels, so the relationship between the device and the drone is as follows:
[0024]
[0025] Step 1.3: Abstract the constraints that the system needs to meet
[0026] To ensure that the drone array covers all data source devices, the following constraints must be met:
[0027]
[0028] Step 1.4: Quantify the optimization goal for this problem
[0029] In order to quantify the energy consumption of the UAV when receiving data, we define E ij The energy consumption when the i-th data source device transmits data to the j-th drone is calculated as follows:
[0030]
[0031] Among them, the channel power gain at the reference distance d0 = 1m is defined as h0, and the Gaussian noise power is defined as σ 2 , the system bandwidth is defined as B, the transmission power and data volume of the i-th device are defined as p i and D i , the transmission rate between the i-th device and the j-th drone is defined as r ij ;
[0032] The total energy consumption of the drone array is calculated as follows:
[0033]
[0034] Among them, the no-load energy consumption and hovering energy consumption of the jth UAV are and The no-load energy consumption of the drone is defined as a fixed value, and the hovering energy consumption of the jth drone is calculated as follows:
[0035]
[0036] The drone array will land only after collecting all the data. The hovering time of the i-th drone is therefore The calculation method is as follows:
[0037]
[0038] Step 1.5: Get the problem model.
[0039] The problem model in step 1.5 is defined as:
[0040]
[0041]
[0042] Where X = {X j |j=1,...,m},Y={Y j |j=1,...,m},U={u j |j=1,...,m}, the weight factor for balancing human-machine energy consumption and equipment energy consumption is defined as X j The lower and upper limits of X are defined as min and X max , Y j The lower and upper limits of Y are defined as min and Y max In the optimization problem defined by this model, X and Y are two sets of continuous variables, and U is a set of integer variables.
[0043] Step 2 is implemented as follows:
[0044] Step 2.1, randomly initialize two populations Pop1 and Pop2 consisting of N individuals, where the integer decision variable part of Pop2 is fixed to 1, indicating that all drones in the system participate in data collection;
[0045] Step 2.2, calculate the objective functions F_Pop1 and F_Pop2 respectively. The calculation method of F_Pop1 is:
[0046]
[0047] The calculation method of F_Pop2 is the same as formula (12) but the integer variable is fixed, that is:
[0048]
[0049] Step 2.3, calculate the overall constraint violation CV(X) of the solution in Pop2:
[0050]
[0051]
[0052] Where p is the inequality constraint g j (X), q is the equality constraint h j (X), since solving equality constraints is more difficult than solving inequality constraints, a very small relaxation factor δ is used to relax equality constraints.
[0053] Step 3 is implemented as follows:
[0054] Step 3.1, select better individuals for reproduction using tournament selection method according to F_Pop1;
[0055] Step 3.2, select better individuals according to F_Pop2 to perform crossover mutation operations to generate new offspring OffSpring2 and repair infeasible solutions in the offspring, evaluate the fitness of the new individual OffSpring2 to obtain F_OffSpring2, calculate the overall constraint violation CV(OffSpring2) of the solution, merge the parent and offspring, and according to the feasibility rule, retain the dominant solution to the next generation and continue to record the infeasible solutions with better target values among the eliminated individuals as the surviving solutions SS to achieve independent evolution of Pop2;
[0056] Step 3.3, monitor the fitness value of Pop2. If the fitness value and constraint violation degree of the population for several consecutive generations are less than the set threshold, it is considered that it has converged to the optimal solution without integer restrictions, recorded as best_solution2. The convergence detection method is shown in the following formula:
[0057] (f′ best -f″ best )≤0and(G′ best -G″ best )≤0 (16)
[0058] Among them, f′ best and G′ best Respectively represent the optimal objective function value and the minimum constraint violation of the current generation, f″ best and G″ best represents the optimal objective function value and the lowest constraint violation found.
[0059] Step 3.1 is implemented as follows:
[0060] Step 3.1.1: For each individual x i , using the mutation operators DE / current-to-rand / 1 and DE / rand-to-best / 1 based on the differential evolution algorithm to generate the offspring mutant individual v i :
[0061] v i =x i +rand×(x r1 -x i )+F×(x r2 -x r3 ) (17)
[0062] v i =x r1 +rand×(x best -x r1 )+F×(x r2 -x r3 ) (18)
[0063] Among them, r1, r2, and r3 are three different integers randomly selected from [1, N], and x best is the best individual in the population, rand is a uniformly distributed random number selected from [0,1], and F is the scaling factor;
[0064] Step 3.1.2: Change the mutant individual v i Individual x of the same parent i Crossover generation offspring experimental vector u i,j :
[0065]
[0066] Among them, i∈{1, 2, ..., N}, j∈{1, 2, ..., D}, cross control parameter CR∈[0,1], rand j is a random number uniformly distributed between [0,1], j rand is an integer randomly selected from [1, D];
[0067] Step 3.1.3: Round off the offspring u generated by the crossover mutation operation i,j The integer variable is used to repair the out-of-bounds individuals in the offspring using the reflection repair strategy, as follows:
[0068]
[0069] For individuals that still violate the constraints after boundary correction, a random repair strategy is adopted to reallocate them to random positions within the boundary, as shown in the following formula:
[0070] u i,j =low+rand(0,1)×(up-low), if u i,j > up or u i,j <low (21)
[0071] The repaired new individual u i,j Denoted as OffSpring1;
[0072] Step 3.1.4, evaluate the fitness value of the new individual OffSpring1 to obtain F_OffSpring1, merge the parent and offspring, and retain the top N individuals with better fitness values through random sorting to achieve independent evolution of Pop1.
[0073] Step 4 is implemented according to the following steps:
[0074] Step 4.1, sort the surviving solutions SS recorded in step 3.2 according to the target value, select the first N-1 solutions and the converged solution best_solution2 as the new Pop2, if the SS is less than N-1, randomly generate solutions as supplements;
[0075] Step 4.2: Based on step 4.1, fix the infeasible solution and complete the fitness evaluation.
[0076] Step 5 is implemented according to the following steps:
[0077] Step 5.1, collect and organize the population data generated in steps 3 and 4, including individual genetic information, fitness values, constraint satisfaction, etc. as data training large model;
[0078] Step 5.2: Sort the target values and constraint violation degrees of the individuals in the two populations, and select the top N individuals with the best target values and constraint violation degrees in the two populations;
[0079] Step 5.3, integrate the integer decision vectors of the first N individuals with better target values and constraint violations in Pop2 into the integer decision vector position of the A-type solution, and integrate the continuous decision vectors of the first N individuals with better target values and constraint violations in Pop1 into the continuous decision vector position of the B-type solution;
[0080] Step 5.4: Perform crossover mutation on the Class A solution and the top N individuals with better target values and constraint violations in Pop2 to retain the top N better solutions; perform crossover mutation on the Class B solution and the top N individuals with better target values and constraint violations in Pop1 to retain the top N better solutions, and repeat step 5.
[0081] In step 6, determine whether the termination condition of Pop1 is met; if not, the large model continues to receive information from the current two populations to dynamically update and optimize the large model; if the termination condition of Pop1 is met, return the final population information of the large model.
[0082] The beneficial effects of the present invention are:
[0083] The data collection array layout method of the present invention driven by a large model and a multi-population mixed integer optimization introduces multiple populations and a large model into the mixed integer programming problem to solve the problems of insufficient operation coverage and large data transmission delay caused by the high average energy consumption of the multi-UAV data collection array layout provided by the prior art, and utilizes the cooperation between the population without integer restrictions and the original population to improve the quality of the population, and improves the quality and diversity of the solution through knowledge sharing and co-evolution; utilizes a large model to learn and predict the performance indicators of the UAV under different layouts to guide the knowledge transfer between the two populations, ensuring that the population can maintain efficient evolutionary capabilities in environmental changes; adopts the solution with positive knowledge eliminated in the environmental selection process as the surviving solution, and the convergent solution of the population without integer restrictions as the preferred gene to form a preferred population, which further improves the optimization effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0084] Figure 1 is a flow chart of a data collection array layout method for large model driven multi-population mixed integer optimization of the present invention;
[0085] Figure 2 It is a schematic diagram of the interaction of a large model in the data collection array layout method of the present invention driven by a large model for multi-population mixed integer optimization;
[0086] Figure 3 It is a schematic diagram of the multi-UAV data collection array layout in the data collection array layout method of the large model driven multi-population mixed integer optimization of the present invention. DETAILED DESCRIPTION
[0087] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0088] Example 1
[0089] The present invention provides a large model driven multi-population mixed integer optimized data collection array layout method, such as Figure 1 As shown, the specific implementation steps are as follows:
[0090] Step 1: Model the multi-UAV data collection array, abstract the system layout optimization model as the optimization problem to be solved by the algorithm, and determine the optimization goal and objective function;
[0091] Step 2: Randomly initialize two populations, Pop1 is used to process tasks with integer restrictions and general constraints, and Pop2 processes tasks that do not contain integer restrictions but only general constraints and evaluate the scalar fitness value with their respective objective functions;
[0092] Step 3: Evolve the two populations independently. Before Pop2 converges to a promising feasible region, record the solutions with good target values but infeasible that are eliminated during the evolution process as surviving solutions.
[0093] Step 4: Use the surviving solution and the converged solution of population 2 as the preferred gene to reinitialize population 2 and obtain a preferred population in a promising area;
[0094] Step 5: The large model drives knowledge transfer, learning and predicting the individuals in the two populations that can provide better target values, and integrating the continuous decision variables of the high-quality individuals in population 2 into the high-quality individuals in population 1 to guide evolution;
[0095] Step 6: Iterate and evolve, dynamically update and optimize the large model, and output the converged solution of population 1 to obtain the optimal solution for the system layout.
[0096] Example 2
[0097] The present invention provides a large model driven multi-population mixed integer optimized data collection array layout method, such as Figure 3 As shown, step 1 is specifically implemented according to the following steps:
[0098] Step 1.1: Abstract the distance calculation method between the data source device and the drone
[0099] Considering the resource consumption of the entire system, not all m drones will participate in the data collection process. Therefore, the variable u is used j ∈{0, 1}(j=1, ..., N) to indicate whether the j-th UAV is used to collect data. If the j-th UAV is used to collect data, then u j =1; otherwise u j = 0. For each data source device, fix its coordinates at (x i ,y i , 0)(i=1,...,γ), the height of the drone from the ground is defined as H, and the coordinates of the jth drone are (X j , Y j ,H), the distance calculation method from the i-th IoT device to the j-th drone is as follows:
[0100]
[0101] Step 1.2: Abstract the relationship model between the data source device and the drone
[0102] The relationship between the i-th IoT device and the j-th drone is defined as a ij , if a ij =1, it means that the i-th data source device sends data to the j-th UAV; otherwise, a ij =0; In the current array, each data source device selects the nearest drone to send data, so the relationship between the device and the drone is as follows:
[0103]
[0104] Each data source device sends data to only one drone, so the relationship between the device and the drone is as follows:
[0105]
[0106] Due to the limitation of data transmission bandwidth, each drone has at most N data receiving channels, so the relationship between the device and the drone is as follows:
[0107]
[0108] Step 1.3: Abstract the constraints that the system needs to meet;
[0109] To ensure that the drone array covers all data source devices, the following constraints must be met:
[0110]
[0111] Step 1.4: quantify the optimization goal under this problem;
[0112] In order to quantify the energy consumption of the UAV when receiving data, we define E ij The energy consumption when the i-th data source device transmits data to the j-th drone is calculated as follows:
[0113]
[0114] Among them, the channel power gain at the reference distance d0 = 1m is defined as h0, and the Gaussian noise power is defined as σ 2 , the system bandwidth is defined as B, the transmission power and data volume of the i-th device are defined as p i and D i , the transmission rate between the i-th device and the j-th drone is defined as r ij ;
[0115] The total energy consumption of the drone array is calculated as follows:
[0116]
[0117] Among them, the no-load energy consumption and hovering energy consumption of the jth UAV are and The no-load energy consumption of the drone is defined as a fixed value, and the hovering energy consumption of the jth drone is calculated as follows:
[0118]
[0119] The drone array will land only after collecting all the data. The hovering time of the jth drone is therefore The calculation method is as follows:
[0120]
[0121] Step 1.5: Get the problem model;
[0122] The layout optimization problem model of this system is defined as follows:
[0123]
[0124] Where X = {X j |j=1,...,m},Y={Y j |j=1,...,m},U={u j |j=1,...,m}, the weight factor for balancing the energy consumption of the drone and the energy consumption of the equipment is defined as X j The lower and upper limits of X are defined as min and X max , Y j The lower and upper limits of Y are defined as min and Y max In the optimization problem defined by this model, X and Y are two sets of continuous variables, and U is a set of integer variables. It is a typical mixed integer programming problem.
[0125] Example 3
[0126] The data collection array layout method of the present invention driven by a large model and multiple population mixed integer optimization, wherein step 2 is specifically implemented according to the following steps:
[0127] Step 2.1, randomly initialize two populations Pop1 and Pop2 consisting of N individuals, where the integer decision variable part of Pop2 is fixed to 1, indicating that all drones in the system participate in data collection;
[0128] Step 2.2, calculate the objective functions F_Pop1 and F_Pop2 respectively. The calculation method of F_Pop1 is:
[0129]
[0130] The calculation method of F_Pop2 is the same as formula (12) but the integer variable is fixed, that is:
[0131]
[0132] Step 2.3, calculate the overall constraint violation CV(X) of the solution in Pop2:
[0133]
[0134] Where p is the inequality constraint g j (X), q is the equality constraint h j (X) Since solving equality constraints is more difficult than solving inequality constraints, a very small relaxation factor δ (usually 1e-6) is used to relax equality constraints;
[0135] Step 3 is implemented as follows:
[0136] Step 3.1, select better individuals for reproduction using tournament selection method according to F_Pop1;
[0137] Step 3.1 is implemented as follows:
[0138] Step 3.1.1: For each individual x i , using the mutation operators DE / current-to-rand / 1 and DE / rand-to-best / 1 based on the differential evolution algorithm to generate the offspring mutant individual v i :
[0139] v i =x i +rand×(x r1 -x i )+F×(x r2 -x r3 ) (17)
[0140] v i =x r1 +rand×(x best -x r1 )+F×(x r2 -x r3 ) (18)
[0141] Among them, r1, r2, and r3 are three different integers randomly selected from [1, N], and x best is the best individual in the population, rand is a uniformly distributed random number selected from [0, 1], and F is the scaling factor;
[0142] Step 3.1.2: Change the mutant individual v i Individual x of the same parent i Crossover generation offspring experimental vector u i,j :
[0143]
[0144] Among them, i∈{1, 2, ..., N}, j∈{1, 2, ..., D}, cross control parameter CR∈[0, 1], rand j is a random number uniformly distributed between [0, 1], j rand is an integer randomly selected from [1, D];
[0145] Step 3.1.3: Round off the offspring u generated by the crossover mutation operation i,j The integer variable is used to repair the out-of-bounds individuals in the offspring using the reflection repair strategy, as follows:
[0146]
[0147] For individuals that still violate the constraints after boundary correction, a random repair strategy is adopted to reallocate them to random positions within the boundary, as shown in the following formula:
[0148] u i,j =low+rand(0,1)×(up-low), if u i,j > up or u i,j <low (21)
[0149] The repaired new individual u i,j Denoted as OffSpring1;
[0150] Step 3.1.4, evaluate the fitness value of the new individual OffSpring1 to obtain F_OffSpring1, merge the parent and offspring, and retain the top N individuals with better fitness values through random sorting to achieve independent evolution of Pop1.
[0151] Step 3.2, select better individuals according to F_Pop2 to perform crossover mutation operations to generate new offspring OffSpring2 and repair infeasible solutions in the offspring, evaluate the fitness of the new individual OffSpring2 to obtain F_OffSpring2, calculate the overall constraint violation CV(O□□□□□ing2) of the solution, merge the parent and offspring, and according to the feasibility rule, retain the dominant solution to the next generation and continue to record the infeasible solutions with better target values among the eliminated individuals as the surviving solutions SS to achieve independent evolution of Pop2;
[0152] Step 3.3, monitor the fitness value of Pop2. If the fitness value and constraint violation degree of the population for several consecutive generations are less than the set threshold, it is considered that it has converged to the optimal solution without integer restrictions, recorded as best_solution2. The convergence detection method is shown in the following formula:
[0153] (f′ best -f′ best )≤0and(G′ best -G″ best )≤0 (16)
[0154] Among them, f′ best and G′ best Respectively represent the optimal objective function value and the minimum constraint violation of the current generation, f″ best and G″ best represents the optimal objective function value and the lowest constraint violation found.
[0155] Step 4 is implemented according to the following steps:
[0156] Step 4.1, sort the surviving solutions SS recorded in step 3.2 according to the target value, select the first N-1 solutions and the converged solution best_solution2 as the new Pop2, if the SS is less than N-1, randomly generate solutions as supplements;
[0157] Step 4.2: Based on step 4.1, fix the infeasible solution and complete the fitness evaluation.
[0158] like Figure 2 As shown, step 5 is specifically implemented according to the following steps:
[0159] Step 5.1: Collect and organize the population data generated in steps 3 and 4, including individual genetic information, fitness values, constraint satisfaction, etc., as data to train the large model. The following is the interaction strategy between the population and the large model:
[0160]
[0161]
[0162] Step 5.2: Sort the target values and constraint violation degrees of the individuals in the two populations, and select the top N individuals with the best target values and constraint violation degrees in the two populations;
[0163] Step 5.3, integrate the integer decision vectors of the first N individuals with better target values and constraint violations in Pop2 into the integer decision vector position of the A-type solution, and integrate the continuous decision vectors of the first N individuals with better target values and constraint violations in Pop1 into the continuous decision vector position of the B-type solution;
[0164] Step 5.4: Perform crossover mutation on the Class A solution and the top N individuals with better target values and constraint violations in Pop2 to retain the top N better solutions; perform crossover mutation on the Class B solution and the top N individuals with better target values and constraint violations in Pop1 to retain the top N better solutions, and repeat step 5.
[0165] In step 6, determine whether the termination condition of Pop1 is met; if not, the large model continues to receive information from the current two populations to dynamically update and optimize the large model; step 6.2, if the termination condition of Pop1 is met, the final population information returned to the large model is the optimal data collection array layout plan, and the optimal solution of Pop1 is output, which is the lowest energy consumption of the layout.
[0166] Through the above methods, the data acquisition array layout optimization method driven by a large model of the present invention for mixed integer optimization of multiple populations introduces multiple populations and large models into the mixed integer programming problem to solve the problems of insufficient operation coverage and large data transmission delay caused by the high average energy consumption of the multi-UAV-assisted data acquisition system layout provided by the prior art. Specifically, the cooperation between the population without integer restrictions and the original population is used to improve the quality of the population, and the quality and diversity of the solution are improved through knowledge sharing and co-evolution; the large model is used to learn and predict the performance indicators of the UAV under different paths to guide the knowledge transfer between the two populations, ensuring that the population can maintain efficient evolutionary capabilities in environmental changes; the solution with positive knowledge eliminated in the environmental selection process is used as the surviving solution, and the convergent solution of the population without integer restrictions is used as the preferred gene to form a preferred population, which further improves the optimization effect.
Claims
1. A data collection array layout method for large model driven multi-population mixed integer optimization, characterized in that: Follow the steps below to implement it: Step 1: Model the multi-UAV-assisted IoT data collection system, abstract the system layout optimization model, and determine the optimization objectives and objective functions; Step 2, randomly initialize two populations; Step 3, evolve the two populations independently. Before population 2 converges to a promising feasible region, record the solutions with good target values but infeasible that are eliminated during the evolution process as surviving solutions. Step 4, reinitialize population 2 using the surviving solution and the converged solution of population 2 as the preferred gene; Step 5: The large model drives knowledge transfer, learning and predicting the individuals in the two populations that can provide better target values, and integrating the continuous decision variables of the high-quality individuals in population 2 into the high-quality individuals in population 1 to guide evolution; Step 6, iterative evolution, dynamic update and optimization of the large model, and output of the population 1 converged solution.
2. The data collection array layout method for large model driven multi-population mixed integer optimization according to claim 1, characterized in that: The step 1 is specifically implemented according to the following steps: Step 1.1: Abstract the distance calculation method between the data source device and the drone Using the variable u j ∈{0,1}(j=1,...,) indicates whether the jth UAV is used to collect data. If the jth UAV is used to collect data, then u j =1; otherwise u j =0; for each data source device, fix its coordinates at ( i ,y i ,0)(i=1,...,), the height of the drone from the ground is defined as H, and the coordinates of the jth drone are ( j ,Y j ,H), the distance calculation method from the i-th IoT device to the j-th drone is: Step 1.2: Abstract the relationship model between the data source device and the drone The relationship between the i-th IoT device and the j-th drone is defined as a ij , if a ij =1, it means that the i-th data source device sends data to the j-th UAV; otherwise, a ij =0; In the current array, each data source device selects the nearest drone to send data, so the relationship between the device and the drone is as follows: Each data source device sends data to only one drone, so the relationship between the device and the drone is as follows: Due to the limitation of data transmission bandwidth, each drone has at most N data receiving channels, so the relationship between the device and the drone is as follows: Step 1.3: Abstract the constraints that the system needs to meet To ensure that the drone array covers all data source devices, the following constraints must be met: Step 1.4: Quantify the optimization goal for this problem In order to quantify the energy consumption of the UAV when receiving data, we define E ij The energy consumption when the i-th data source device transmits data to the j-th drone is calculated as follows: Among them, the channel power gain at the reference distance d0 = 1m is defined as h0, and the Gaussian noise power is defined as σ 2 , the system bandwidth is defined as B, the transmission power and data volume of the i-th device are defined as p i and D i , the transmission rate between the i-th device and the j-th drone is defined as r ij ; The total energy consumption of the drone array is calculated as follows: Among them, the no-load energy consumption and hovering energy consumption of the jth UAV are and The no-load energy consumption of the drone is defined as a fixed value, and the hovering energy consumption of the jth drone is calculated as follows: The drone array will land only after collecting all the data. The hovering time of the jth drone is therefore The calculation method is as follows: Step 1.5: Get the problem model.
3. The data collection array layout method for large model driven multi-population mixed integer optimization according to claim 2, characterized in that: The problem model in step 1.5 is defined as: Where X = {X j |j=1,…,m},Y={Y j |j=1,…,m},U={u j |=1,…,}, the weight factor for balancing human-machine energy consumption and equipment energy consumption is defined as X j The lower and upper limits are defined as min and max ,Y j The lower and upper limits of Y are defined as min and Y max In the optimization problem defined by this model, X and Y are two sets of continuous variables, and U is a set of integer variables.
4. The data collection array layout method for large model driven multi-population mixed integer optimization according to claim 1, characterized in that: The step 2 is specifically implemented according to the following steps: Step 2.1, randomly initialize two populations Pop1 and Pop2 consisting of N individuals, where the integer decision variable part of Pop2 is fixed to 1, indicating that all drones in the system participate in data collection; Step 2.2, calculate the objective functions F_Pop1 and F_Pop2 respectively. The calculation method of F_Pop1 is: The calculation method of F_Pop2 is the same as formula (12) but the integer variable is fixed, that is: Step 2.3, calculate the overall constraint violation CV(X) of the solution in pop2: Where p is the inequality constraint g j (X), q is the equality constraint h j (X), since solving equality constraints is more difficult than solving inequality constraints, a very small relaxation factor δ is used to relax equality constraints.
5. The data collection array layout method for large model driven multi-population mixed integer optimization according to claim 1, characterized in that: The step 3 is specifically implemented according to the following steps: Step 3.1, select better individuals for reproduction using tournament selection method according to F_Pop1; Step 3.2: Select better individuals according to F_Pop2 and perform crossover mutation operations to generate new offspring OffSpring2 and repair infeasible solutions in the offspring. Perform fitness evaluation on the new individual OffSpring2 to obtain F_OffSpring2. Calculate the overall constraint violation CV(2) of the solution to merge the parent and offspring. According to the feasibility rule, retain the dominant solution to the next generation and continue to record the infeasible solutions with better target values among the eliminated individuals as the surviving solutions SS to achieve independent evolution of Pop2. Step 3.3, monitor the fitness value of Pop2. If the fitness value and constraint violation degree of the population for several consecutive generations are less than the set threshold, it is considered that it has converged to the optimal solution without integer restrictions, recorded as best_2. The convergence detection method is shown in the following formula: (f′best-b″ best )≤0 and(G′ best -b′ best )≤0 (16) Among them, f′ best and G′ best Respectively represent the optimal objective function value and the minimum constraint violation of the current generation, f″ best and G″ best represents the optimal objective function value and the lowest constraint violation found.
6. The data collection array layout method for large model driven multi-population mixed integer optimization according to claim 5, characterized in that: The step 3.1 is specifically implemented according to the following steps: Step 3.1.1: For each individual x i , using the mutation operator based on differential evolution algorithm DE / current-to-rand / 1 and DE / rand-to-best / 1 generate offspring mutation individuals v i : v i =x i +rand×(x r1 -x i )+F×(x r2 -x r3 ) (17) v i =x r1 +rand×(x best -x r1 )+F×(x r2 -x r3 ) (18) Among them, r1, r2, and r3 are three different integers randomly selected from [1, N], and x best is the best individual in the population, rand is a uniformly distributed random number selected from [0,1], and F is the scaling factor; Step 3.1.2: Change the mutant individual v i Individual x of the same parent i Crossover generation offspring experimental vector u i,j : Among them, i∈{1,2,…,N},j∈{1,2,…,D}, cross control parameter CR∈[0,1], rand j is a random number uniformly distributed between [0,1], j rand is an integer randomly selected from [1,D]; Step 3.1.3: Round off the offspring u generated by the crossover mutation operation i,j The integer variable is used to repair the out-of-bounds individuals in the offspring using the reflection repair strategy, as follows: For individuals that still violate the constraints after boundary correction, a random repair strategy is adopted to reallocate them to random positions within the boundary, as shown in the following formula: u i,j = low + rand(0,1)×(up - low), if u i,j > up or u i,j < low (21) The newly repaired individual u i,j is denoted as OffSpring1; Step 3.1.4, evaluate the fitness value of the new individual OffSpring1 to obtain F_OffSpring1, merge the parent and offspring, and retain the first N individuals with better fitness values by random sorting to achieve Independent evolution of Pop1.
7. The data collection array layout method for large model driven multi-population mixed integer optimization according to claim 1, characterized in that: The step 4 is specifically implemented according to the following steps: Step 4.1, sort the surviving solutions SS recorded in step 3.2 according to the target value, select the first N-1 solutions and the converged solution best_2 as the new Pop2, if SS is less than N-1, randomly generate solutions as supplements; Step 4.2: Based on step 4.1, fix the infeasible solution and complete the fitness evaluation.
8. The data collection array layout method for large model driven multi-population mixed integer optimization according to claim 1, characterized in that: The step 5 is specifically implemented according to the following steps: Step 5.1, collect and organize the population data generated in steps 3 and 4, including individual genetic information, fitness values, constraint satisfaction, etc. as data training large model; Step 5.2: Sort the target values and constraint violation degrees of the individuals in the two populations, and select the top N individuals with the best target values and constraint violation degrees in the two populations; Step 5.3, integrate the integer decision vectors of the first N individuals with better target values and constraint violations in Pop2 into the integer decision vector position of the A-type solution, and integrate the continuous decision vectors of the first N individuals with better target values and constraint violations in Pop1 into the continuous decision vector position of the B-type solution; Step 5.4: Perform crossover mutation on the Class A solution and the top N individuals with better target values and constraint violations in Pop2 to retain the top N better solutions; perform crossover mutation on the Class B solution and the top N individuals with better target values and constraint violations in Pop1 to retain the top N better solutions, and repeat step 5.
9. The data collection array layout method for large model driven multi-population mixed integer optimization according to claim 1, characterized in that: In step 6, it is determined whether the termination condition of Pop1 is met; if not, the large model continues to receive information of the current two populations to dynamically update and optimize the large model; if the termination condition of Pop1 is met, the final population information of the large model is returned.
Citation Information
Cited By
Multi-objective optimization method for performance of steel wheel
CN120068269A
