Power transmission line unmanned aerial vehicle full-coverage autonomous inspection scheduling method based on leapfrog algorithm

Abstract

The invention relates to a power transmission line unmanned aerial vehicle full-coverage autonomous inspection scheduling method based on a leapfrog algorithm. The method comprises the steps of obtaining a tower number and an inspection task load; determining the number and number of the unmanned aerial vehicles and the poor inspection quality degree of the unmanned aerial vehicles; establishing an optimal scheduling model; designing a double-layer coding mechanism; the shuffled frog leaping algorithm is improved; and solving the optimal scheduling model to obtain an optimal inspection scheduling scheme. According to the method, the characteristics of the unmanned aerial vehicle full-coverage autonomous inspection scheduling problem of the power transmission line are combined, a dual optimization target is constructed, an optimization model of unmanned aerial vehicle full-coverage autonomous inspection of the power transmission line is established, the optimization model is solved by adopting a shuffled frog leaping algorithm, and an optimal inspection scheduling scheme is obtained. Power transmission line inspection is executed according to the optimal scheduling scheme obtained by the method, the operation time of the unmanned aerial vehicle can be greatly saved, the power transmission line inspection efficiency is effectively improved, the inspection quality is guaranteed, and powerful support is provided for power grid operation and maintenance work.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of power transmission line maintenance, and particularly relates to a power transmission line unmanned aerial vehicle (UAV) full-coverage autonomous inspection scheduling method based on a frog leap algorithm. BACKGROUND

[0002] In recent years, with the rapid development of small and medium-sized civilian unmanned aerial vehicle technology and the popularization of unmanned aerial vehicle driving skills, unmanned aerial vehicles have been widely applied to power transmission line inspection work. Unmanned aerial vehicle inspection has entered a rapid development stage, and at present, autonomous and refined inspection has been widely applied in power grid companies, which has brought great results to power transmission line operation and maintenance work. However, autonomous and refined inspection is mainly aimed at towers and their accessories of power transmission lines, but its inspection effect on ground wires and channels is limited. In order to continuously promote the large-scale application of unmanned aerial vehicles in power transmission lines and improve operation and maintenance quality, it will become a trend to simultaneously carry out autonomous inspection of ground wires and channels.

[0003] Power transmission line unmanned aerial vehicle full-coverage autonomous inspection refers to a comprehensive inspection combining three types of tower autonomous refined inspection, ground wire autonomous inspection and channel autonomous inspection, which covers all the contents required to be inspected of power transmission line equipment, accessories and channels. Power transmission line unmanned aerial vehicle full-coverage autonomous inspection is an improvement and supplement to the refined inspection mode, and its characteristics are to realize full autonomy of inspection, and unmanned aerial vehicles automatically navigate and inspect according to the flight route; the inspection object and content range become comprehensive.

[0004] When performing unmanned aerial vehicle full-coverage autonomous inspection, in the face of numerous task towers in the region, if the number of towers is increased or decreased according to the direction, and multiple unmanned aerial vehicles are manually dispatched to carry out work according to the on-site situation, there is great blindness, low work efficiency and cannot guarantee the inspection quality. SUMMARY

[0005] The purpose of the present application is to provide a power transmission line unmanned aerial vehicle full-coverage autonomous inspection scheduling method based on a frog leap algorithm, which solves the above problems. The inspection task of the unmanned aerial vehicle includes three types of power transmission line towers, ground wires and channels. For the unmanned aerial vehicle inspection task scheduling problem, the maximum inspection completion time and the inspection quality are taken as the optimization target together, an optimal scheduling model for unmanned aerial vehicle inspection scheduling is established, the hybrid frog leap algorithm is deeply improved, the optimal solution of the unmanned aerial vehicle inspection task scheduling problem is obtained, and the digital transformation and intelligent upgrading of unmanned aerial vehicle inspection are promoted.

[0006] The technical scheme of the present application is a power transmission line unmanned aerial vehicle full-coverage autonomous inspection scheduling method based on a frog leap algorithm, which comprises the following steps:

[0007] Step 1: obtaining the numbers of power transmission line towers to be inspected in an inspection area and the task quantities of different types of tower inspection tasks;

[0008] Step 2: determine the number of unmanned aerial vehicles for tower inspection, the number of unmanned aerial vehicles available for each type of inspection task and the specific unmanned aerial vehicle number, the inspection speed of each unmanned aerial vehicle and the corresponding inspection quality bad degree;

[0009] Step 3: establish an optimal scheduling model for unmanned aerial vehicle full coverage autonomous inspection scheduling;

[0010] Step 3.1: determine the optimization objective of the optimal scheduling model;

[0011] Step 3.1.1: construct the first objective function of the optimal scheduling model as the minimum value of the maximum completion time of the unmanned aerial vehicle completing full coverage autonomous inspection;

[0012] Step 3.1.2: construct the second objective function of the optimal scheduling model as the minimum value of the average of the unmanned aerial vehicle inspection quality bad degree;

[0013] Step 3.2: establish the constraint condition of the optimal scheduling model;

[0014] Step 4: develop a double-layer coding mechanism containing task tower sorting and unmanned aerial vehicle inspection speed selection information;

[0015] Step 5: improve the hybrid frog leap algorithm in combination with the characteristics of the optimal scheduling model, and propose an external archive maintenance method based on iteration repetition rate;

[0016] Step 6: use the improved hybrid frog leap algorithm obtained in step 5 to solve the optimal scheduling model in step 3 to obtain the optimal inspection scheduling scheme.

[0017] Further, in step 1, the different types of tower inspection tasks include tower autonomous fine inspection task, ground wire autonomous inspection task and channel autonomous inspection task.

[0018] In step 2, n represents the number of inspection task towers in the area, x represents the number of the tower to be inspected, U represents the set of unmanned aerial vehicles, UAV k represents the set of unmanned aerial vehicles for the kth type of inspection task, k represents the type of inspection, i.e. the inspection stage, k=1 represents the tower autonomous fine inspection task, k=2 represents the ground wire autonomous inspection task, and k=3 represents the channel autonomous inspection task; u k represents the number of unmanned aerial vehicles for the kth type of inspection task; represents the number of the j k th unmanned aerial vehicle for the kth type of inspection task, represents the k x th type of inspection task of the i th tower, represents the inspection speed of the unmanned aerial vehicle represents the inspection task amount of the inspection task , V represents the inspection speed of the unmanned aerial vehicle task, represents the unmanned aerial vehicle , and T x represents the time required for the i th tower to complete the k th inspection task, represents the inspection start time of the inspection task, represents the inspection completion time of the inspection task, F max represents the maximum completion time, represents the unmanned aerial vehicle , and Q represents the badness degree of the inspection quality of the unmanned aerial vehicle

[0019] Set: the unmanned aerial vehicles in the set U of unmanned aerial vehicles can be used for inspection work, a single unmanned aerial vehicle can only perform one type of tower inspection task at the same time, and cannot be interrupted; the number of unmanned aerial vehicles used for the k k th type of inspection task u

[0020] i x =1, 2,..., n, i′ x =1, 2,..., n (1)

[0021] j k =1, 2,..., u k (2)

[0022] k=1 or 2 or 3 (3)

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046] Equation (6) and equation (7) represent the binary decision of tower inspection; equation (8) and equation (9) represent the priority relationship between towers in the same inspection stage; equation (10) and equation (11) represent the priority relationship between different inspection stages of the same tower; equation (12) defines the maximum completion time as the time when the last inspection is completed; equation (13) ensures that the start time of the first inspection stage is greater than or equal to zero; equation (14) ensures that the unmanned aerial vehicle inspects each tower at the selected inspection speed in any inspection stage; equation (15) calculates the completion time of any inspection stage; equation (16) formulates constraints to ensure that the same tower must complete the current inspection stage before proceeding to the next stage of inspection; equation (17) and equation (18) ensure the constraint relationship between different towers in the same inspection stage; equation (19) and equation (20) ensure the constraint relationship between different inspection stages of the same tower.

[0047] In step 3.1.1, the minimum value of the maximum completion time of the full coverage autonomous inspection is:

[0048]

[0049] In the formula, f1 represents the first objective function; represents the completion time of the i x th tower in the k th inspection stage.

[0050] In step 3.1.2, the average of the minimization of the badness degree of the inspection quality is calculated:

[0051]

[0052] In the formula, f2 represents the second objective function.

[0053] In step 5, the hybrid frog leap algorithm is improved to obtain an improved hybrid frog leap algorithm, including the following steps:

[0054] (I) Initialize the population and the external archive;

[0055] Initialize the frog population P = {O1, O2,..., ON}, O N p = 1, 2,..., N, each frog in the frog population corresponds to a feasible solution of the optimal scheduling model, and O p = [O p 1, O pα 2,..., O pβ] N} represents the first objective function value of the p T th frog, O pα 2 represents the second objective function value of the p pβ th frog,

[0056] The frog population initialization parameters are generated by a random number generation method to ensure the dispersion of the initialized frog population individuals; and the external archive is initialized.

[0057] (II) Construct a relationship pairing matrix of frog individuals;

[0058] Calculate the variation index value L p for each frog individual O p , and pair two individuals at a time, when the p th frog in the population O 1 is paired with the p th frog O 2, a label value is set according to the size comparison result of the variation index value, if L p1 p2 L p1 p2

[0059] The label value of each pair of individuals is taken as an element in the frog individual relationship pairing matrix, and the frog individual relationship pairing matrix is constructed;

[0060] (iii) constructing and training the radius neighbor classifier RNC;

[0061] The frog individual relationship pairing matrix is used as training data, and the label value in the frog individual relationship pairing matrix is used as the training data label. The radius neighbor classifier RNC is constructed and trained, and the parameter value of the radius threshold Radius is determined through hyperparameter search;

[0062] (iv) dividing the population into multiple levels;

[0063] The population is divided into levels according to the dominance relationship in the population, where the first level is all non-dominated solutions, the second level is solutions dominated by the first level, and so on. The frog population is divided into multiple levels;

[0064] The fast non-dominated sorting algorithm is used to sort the pth frog individual O p The layering parameter R p is assigned, p = 1, 2, … N, frog individuals with the same layering parameter value are divided into the same layer, and the maximum number of layers is Lmax, which represents the number of frog population levels; if R p = l, then O p is divided into the lth layer;

[0065] (v) calculating the crowding degree value;

[0066] For the layered frog population, the crowding degree value C p of the pth frog individual O p and other individuals in the population is calculated, which is used to evaluate the surrounding density of frog individuals in the population;

[0067] For frog individuals with the same layering parameter value in the population, the frog individual with a larger crowding degree value is better;

[0068] Let the crowding degree value of the frog individual with the largest crowding degree value in the lth layer of the frog population be Cmax l , and the crowding degree value of the frog individual with the smallest crowding degree value be Cmin l , l = 1, 2, … Lmax, Lmax represents the number of frog population levels;

[0069] (vi) retaining the best individual in the level;

[0070] The frog individuals in the lth layer of the frog population are sorted according to the crowding degree value from large to small, and the first n l frog individuals in the obtained frog individual sequence are taken as the elite set E, which is placed in the next generation population Q, n lthe number of elite frog individuals of the first layer of the frog population,

[0071]

[0072] (vii) Constructing meme groups;

[0073] The frog individuals in the next generation population Q are sorted according to the crowding degree values and then evenly distributed to m Q different meme groups, m Q represents the number of meme groups;

[0074] (viii) Meme group evolution and individual update;

[0075] The frog individuals in each meme group are subjected to mutation and crossover operations to generate new solutions O new The generated new solutions O new and the old solutions O org in the existing meme groups are paired, and the radius neighbor classifier RNC trained in step (iii) is used to predict the label of the relationship between the frog individuals, if the predicted label value is 1 or 0, it indicates that the new solution O new is superior to or equal to the old solution O org , then the new solution O new replaces the old solution O org , if the predicted label value is -1, it indicates that the new solution O new is inferior to the old solution O org , then a random solution is generated to replace the old solution O org ;

[0076] (ix) External archive maintenance;

[0077] The evolved meme groups are added to the external archive, and if the capacity of the external archive exceeds the upper limit, a screening and deletion maintenance operation is performed;

[0078] (x) Judgment of whether the algorithm termination condition is met;

[0079] A proportion Fr of the solutions in the external archive are randomly extracted to form a new population P', Fr represents the proportion of the extracted solutions in the external archive, and it is judged whether the algorithm termination condition is met, if the termination condition is met, the search is stopped and the non-dominated solutions are returned; otherwise, step (iv) is executed.

[0080] Compared with the prior art, the present application has the following beneficial effects:

[0081] 1) The present application combines the characteristics of the unmanned aerial vehicle full coverage autonomous inspection scheduling problem of the power transmission line, constructs a dual optimization objective of minimizing the maximum inspection completion time and minimizing the poor degree of inspection quality, establishes an optimization model of the unmanned aerial vehicle full coverage autonomous inspection of the power transmission line, and solves the optimization model by using a hybrid frog leap algorithm to obtain an optimal inspection scheduling scheme. The optimal scheduling scheme obtained by the method of the present application can greatly save the unmanned aerial vehicle operation time, effectively improve the power transmission line inspection efficiency, and guarantee the inspection quality, thereby providing strong support for power grid operation and maintenance work.

[0082] 2) The present application improves the hybrid frog leap algorithm by using a multi-objective elite reservation strategy for solving the optimization model of the unmanned aerial vehicle full coverage autonomous inspection of the power transmission line. The improved hybrid frog leap algorithm improves the search speed and accuracy of the hybrid frog leap algorithm by reserving a certain number of non-inferior solution individuals, thereby improving the efficiency of formulating the unmanned aerial vehicle inspection scheduling scheme.

[0083] 3) The present application adopts an external archive maintenance method based on an iteration repetition rate in the improved hybrid frog leap algorithm. The method quickly locates the repeatedly occurring solutions in the iteration of the frog leap algorithm through a counter, performs a differentiated probability deletion operation, improves the diversity and convergence speed of the frog leap algorithm search, and is simple and efficient in process, which is conducive to finding a global optimal solution and avoiding falling into a local optimal solution.

[0084] 4) The present application adopts a double-layer coding mechanism in the feasible solution of the optimization model of the unmanned aerial vehicle inspection scheduling scheme, i.e., the unmanned aerial vehicle full coverage autonomous inspection of the power transmission line, thereby improving the accuracy of the expression of the comprehensive inspection scheduling problem. The double-layer coding more directly represents the constraint relationship, which is conducive to improving the efficiency and accuracy of the subsequent optimization model solution.

[0085] 5) The present application uses artificial intelligence models and methods to realize automatic scheduling optimization of the unmanned aerial vehicle inspection operation, and promotes the digital transformation and intelligent upgrading of the unmanned aerial vehicle inspection. BRIEF DESCRIPTION OF DRAWINGS

[0086] The present application will be further described below in combination with the drawings and examples.

[0087] Figure 1 FIG. 1 is a schematic diagram of the power transmission line unmanned aerial vehicle full coverage autonomous inspection scheduling method of the present application.

[0088] Figure 2 FIG. 2 is a flowchart of generating a scheduling scheme candidate set in the present application.

[0089] Figure 3 FIG. 3 is a Gantt chart of the scheduling scheme of the present application. DETAILED DESCRIPTION

[0090] AsFigure 1 The power transmission line unmanned aerial vehicle full coverage autonomous inspection scheduling method based on the frog algorithm, as shown in the figure, comprises the following steps:

[0091] Step 1: Obtain the number of power transmission line towers to be inspected in the inspection area and the task amount of different types of tower inspection tasks;

[0092] In the embodiment, the different types of tower inspection tasks include tower autonomous fine inspection tasks, ground wire autonomous inspection tasks, and channel autonomous inspection tasks.

[0093] The inspection task amount of the tower to be inspected in the embodiment is shown in Table 1. The task amount is represented by power-time product PTP, and the unit is 60 times watt-second, i.e. 60Ws.

[0094] Table 1 Tower inspection task amount table

[0095] Tower No. Phase 1 Phase 2 Phase 3 1 102 78 110 2 116 48 62 3 52 116 118 4 54 70 62 5 84 96 66

[0096] Step 2: Determine the number of unmanned aerial vehicles for tower inspection, the number of unmanned aerial vehicles available for each type of inspection task and the specific unmanned aerial vehicle number, the inspection speed of each unmanned aerial vehicle, and the corresponding inspection quality bad degree;

[0097] Define n to represent the number of inspection task towers in the area, x to represent the number of towers to be inspected, U to represent the set of unmanned aerial vehicles, UAV k represents the set of unmanned aerial vehicles for the kth type of inspection task, k represents the type of inspection, i.e. the inspection stage, k=1 represents the tower autonomous fine inspection task, k=2 represents the ground wire autonomous inspection task, and k=3 represents the channel autonomous inspection task; u k represents the number of unmanned aerial vehicles for the kth type of inspection task; represents the number of the j k th unmanned aerial vehicle for the kth type of inspection task, represents the kth type of inspection task of the i x th tower, represents the inspection speed of the unmanned aerial vehicle , represents the inspection task amount of the inspection task , V represents the inspection speed set composed of unmanned aerial vehicle task inspection speeds, represents the time required for the unmanned aerial vehicle to complete the kth type of inspection task for the i x th tower, represents the start time of the inspection task , represents the completion time of the inspection task , and F maxdenotes the maximum completion time, denotes the unmanned aerial vehicle at a speed the bad degree of the inspection quality of the inspection.

[0098] The inspection speed of the unmanned aerial vehicle in the embodiment has 0 gear, 1 gear, 2 gear and the like, as shown in Table 2.

[0099] Table 2: Unmanned aerial vehicle inspection speed table

[0100]

[0101] Step 3: Establish an optimal scheduling model for the unmanned aerial vehicle full-coverage autonomous inspection scheduling;

[0102] Step 3.1: Determine the optimization objective of the optimal scheduling model;

[0103] Step 3.1.1: The first objective function of the optimal scheduling model is to solve the minimum value of the maximum completion time of the unmanned aerial vehicle completing the full-coverage autonomous inspection,

[0104]

[0105] In the formula, f1 denotes the first objective function; denotes the completion time of the inspection task of the i x th tower at the kth inspection stage.

[0106] Step 3.1.2: The second objective function of the optimal scheduling model is to solve the minimum value of the average value of the bad degree of the unmanned aerial vehicle inspection quality,

[0107]

[0108] Step 3.2: Establish the constraint condition of the optimal scheduling model;

[0109] It is set that: the unmanned aerial vehicles in the set U of unmanned aerial vehicles can be used for inspection work, a single unmanned aerial vehicle can only perform one kind of tower inspection task at the same time, and cannot be interrupted; the number of unmanned aerial vehicles used for the kth kind of inspection task u k is not less than 1; the task amount of path flight between the take-off point or the end point of the previous inspection task and the power line inspection task point is included in the task amount of the current inspection task;

[0110] i x =1,2,...,n,i′ x =1,2,...,n (1)

[0111] j k =1,2,...,u k (2)​

[0112] k = 1 or 2 or 3 (3)

[0113]

[0114]

[0115]

[0116]

[0117]

[0118]

[0119]

[0120]

[0121]

[0122]

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129]

[0130]

[0131]

[0132]

[0133]

[0134]

[0135]

[0136] Equation (6) and equation (7) represent the binary decision of the tower inspection task; equation (8) and equation (9) represent the priority relationship between the towers in the same inspection stage; equation (10) and equation (11) represent the priority relationship between different inspection stages of the same tower; equation (12) defines the maximum completion time as the time when the last inspection is completed; equation (13) ensures that the start time of the first inspection stage is greater than or equal to zero; equation (14) ensures that the unmanned aerial vehicle inspects each tower at the selected inspection speed in any inspection stage; equation (15) calculates the completion time of any inspection stage; equation (16) formulates a constraint to ensure that the same tower must complete the current inspection task before proceeding to the next stage of inspection; equation (17) and equation (18) ensure the constraint relationship between different towers in the same inspection stage; equation (19) and equation (20) ensure the constraint relationship between different inspection stages of the same tower.

[0137] Step 4: Develop a double-layer encoding mechanism containing task tower sorting and unmanned aerial vehicle inspection speed selection information;

[0138] The first layer of the double-layer encoding of the scheduling scheme is the tower inspection sequence encoding X1X2X3…X n , where i x =1,2,…n represents the tower with inspection number i x , and n is the number of inspection task towers;

[0139] The second layer of the double-layer encoding of the scheduling scheme is the constraint encoding, which uses an n-row and 3-column matrix to represent the unmanned aerial vehicle inspection speed selection information of each tower in different inspection stages,

[0140]

[0141] In the formula, is the unmanned aerial vehicle inspection speed selection matrix of the n-base task tower, i x =1,2,…n,k=1,2,3 represents the unmanned aerial vehicle inspection speed of the i x th tower in the kth inspection stage.

[0142] In the embodiment, the double-layer encoding of the scheduling scheme uses the data structure En, En[1] represents the first layer data of the double-layer encoding, i.e. the tower inspection sequence encoding, En[1]=X1X2X3…X n ;

[0143] En[2] represents the second layer data of the double-layer encoding of the scheduling scheme, i.e. the unmanned aerial vehicle inspection speed selection matrix of the task tower,

[0144] Step 5: Based on the characteristics of the optimal scheduling model, the hybrid frog leap algorithm is improved, and an external archive maintenance method based on iteration repetition rate is proposed;

[0145] (I) Initialize the population and external archive;

[0146] Initialize the frog population P = {O1, O2,..., ON}, N represents the number of frogs in the population, and each frog corresponds to a feasible solution of the optimal scheduling model. N}, O p p = 1, 2,..., N represents the pth frog, and each frog in the frog population corresponds to a feasible solution of the optimal scheduling model. O p = [O pα , O pβ] T where O pα represents the first objective function value of the pth frog, O pβ represents the second objective function value of the pth frog,

[0147] The frog population initialization parameters are generated using a random number generation method to ensure the dispersion of the initialized frog population individuals; initialize the external archive;

[0148] (II) Construct the frog individual relationship pairing matrix;

[0149] For each frog individual O p , calculate the variant index value L p , and pair each individual with another individual. When the p1th frog O p1 in the population is paired with the p2th frog , set the label value according to the size comparison result of the variant index value. If record represents the label value of the p1th frog in the population paired with the p2th frog ; if record If record

[0150] The label value of the pair of individuals is used as an element in the frog individual relationship pairing matrix to construct the frog individual relationship pairing matrix.

[0151] (III) Construct and train the radius neighbor classifier RNC, and use the radius neighbor classifier RNC as the comparison operator of the frog individual;

[0152] Use the frog individual relationship pairing matrix as the training data, and use the label value in the frog individual relationship pairing matrix as the training data label to construct the radius neighbor classifier RNC and train it. The parameter value of the radius threshold Radius is determined through hyperparameter search.

[0153] (iv) dividing the population into multiple levels;

[0154] The population is divided into levels according to the dominance relationship in the population, wherein the first level is all non-dominated solutions, the second level is solutions dominated by the first level, and so on, and the frog population is divided into multiple levels;

[0155] The fast non-dominated sorting algorithm is used to sort the pth frog individual O p The hierarchical parameter R p is assigned, p = 1, 2, … N, frog individuals with the same hierarchical parameter value are divided into the same level, the maximum number of levels is Lmax, and Lmax represents the number of levels of the frog population; if R p = l, then O p is divided into the lth level;

[0156] (v) calculating the crowding degree value;

[0157] The crowding degree value C p of the pth frog individual O p and other individuals in the population is calculated, which is used to evaluate the surrounding density of the frog individual in the population;

[0158] For frog individuals with the same hierarchical parameter value in the population, the frog individual with a larger crowding degree value is better;

[0159] Let the crowding degree value of the frog individual with the largest crowding degree value in the lth level of the frog population be Cmax l , the crowding degree value of the frog individual with the smallest crowding degree value be Cmin l , and l = 1, 2, … Lmax, Lmax representing the number of levels of the frog population;

[0160] (vi) retaining the optimal individual in the level;

[0161] The frog individuals in the lth level of the frog population are sorted according to the crowding degree value from large to small, and the first n l frog individuals in the obtained frog individual sequence are taken out as an elite set E, which is placed into the next generation population Q, n l being the number of elite frog individuals in the lth level of the frog population,

[0162]

[0163] (vii) constructing a meme group;

[0164] The frog individuals in the next generation population Q are evenly distributed to m Q different meme groups according to the crowding degree value, m Q representing the number of meme groups;

[0165] (eight) meme group evolution and individual update;

[0166] The frog individuals in each meme group are mutated and crossed to generate new solutions O new The generated new solutions O new and the old solutions O org in the existing meme group are paired, and the label of the relationship between the frog individuals is predicted by the radius neighbor classifier RNC trained in step (three). If the predicted label value is 1 or 0, it means that the new solution O new is better than or equal to the old solution O org , and the new solution O new replaces the old solution O org . If the predicted label value is -1, it means that the new solution O new is worse than the old solution O org , and a random solution replaces the old solution O org .

[0167] (nine) external archive maintenance;

[0168] The entire individual set of the evolved meme group is added to the external archive. If the capacity of the external archive exceeds the upper limit, a maintenance operation of screening and deleting is performed.

[0169] (ten) judgment of whether the algorithm termination condition is met;

[0170] A proportion Fr of the solutions in the external archive is randomly extracted to form a new population P' of the next generation. Fr represents the proportion of the extracted solutions in the external archive. It is judged whether the algorithm termination condition is met. If the termination condition is met, the search is stopped and the non-dominated solution is returned. Otherwise, step (four) is executed.

[0171] Based on the iteration repetition rate, the external archive is maintained, specifically including:

[0172] The non-dominated solutions of the frog individuals O p entering the external archive are set with corresponding counters Counter p , Counter p representing the counter corresponding to the pth frog individual O p , which is used to record the number of times the same non-dominated solution appears in iterations. When the capacity of the external archive reaches a fixed proportion k f of the upper limit, k f represents the proportion corresponding to the capacity upper limit of the external archive; the iteration repetition rate IR p of the non-dominated solution O p is calculated, and elimination is performed according to the corresponding elimination probability EP p .

[0173] IR p = Counterp / Iter now

[0174]

[0175] EP0 = 0, EP1 < EP2 < 1

[0176] In the formula Iter now Indicates the current iteration number; IR p O represents the p-th frog individual. p The iteration repetition rate of non-dominated solutions; EP p O represents the p-th frog individual. p The elimination probability of a non-dominated solution; EP0 represents the elimination probability of a small iteration repetition rate, EP1 represents the elimination probability of a medium-to-high iteration repetition rate, and EP2 represents the elimination probability of a large iteration repetition rate.

[0177] Step 6: Use the improved hybrid frog-jumping algorithm obtained in Step 5 to solve the optimal scheduling model in Step 3, and obtain the optimal inspection scheduling scheme.

[0178] like Figure 2 As shown, the scheduling scheme generation method using two-layer coding includes:

[0179] 1) Generate a random sequence for the inspection towers to obtain the tower inspection sequence code X1X2X3…X n The remainder selection model is used to determine the UAV inspection speed selection matrix for n-base task towers.

[0180] 2) Initialize the inspection phase k, let k = 1;

[0181] 3) Code the tower inspection sequence as X1X2X3…X n The corresponding tower sequence determines the inspection scheduling order of the towers in the task, resulting in the tower sequence for the k-th stage inspection task. Let i x =1;

[0182] 4) Extract the i-th tower sequence from the tower sequence of the k-th stage inspection task. x There are several inspection towers, starting from the one that can perform the inspection task of the kth inspection stage. k The drone with the earliest standby time among the drones is selected as the current i-th drone. x When a tower is performing an inspection task, if there are multiple drones available, one drone is randomly selected, and the drone's inspection speed is determined using a random sampling method.

[0183] The range of the drone's mission inspection speed in the embodiment is [3, 98].

[0184] 5) update the standby time of the selected unmanned aerial vehicle for performing the i x th task tower kth stage inspection task in step 4); update the completion time of the i x th task tower for performing the kth stage inspection task;

[0185] 6) determine whether i x <n is true, if true, set i x =i x +1, execute step 4), otherwise, execute step 7);

[0186] 7) determine whether k<3 is true, if true, determine the k+1th stage tower inspection scheduling order according to the order of the completion time of the kth stage task tower inspection, obtain the tower sequence of the k+1th stage inspection task, set k=k+1, execute step 4), otherwise, execute step 8);

[0187] 8) obtain the candidate scheduling scheme according to the calculation results of steps 1) to 7).

[0188] In the embodiment, the first layer coding and the second layer constraint coding of the double-layer coding of a certain candidate scheduling scheme obtained by using the above scheduling scheme generation method are shown in Table 3 and Table 4, and the inspection speeds of the 1st to 5th towers in the 1st to 3rd stages of the candidate scheduling scheme determined by using the remainder selection model are shown in Table 5.

[0189] A certain candidate scheduling scheme obtained in the embodiment is shown in Table 6. Figure 3

[0190] The pseudo code of the radius neighbor classifier RNC implemented in the embodiment is as follows.

[0191]

[0192]

[0193] The fast non-dominated solution sorting algorithm of the embodiment refers to the non-inferior sorting algorithm disclosed in the book “Multi-objective Intelligent Optimization Algorithm and Its Application” edited by Leideng and Yan Xinping and published by Scientific Publishing in 2009.

[0194] Table 3 first layer coding of double-layer coding of candidate scheduling scheme

[0195] Serial No. 1 2 3 4 5 Tower No. 4 2 5 1 3

[0196] Table 4 second layer coding of double-layer coding of candidate scheduling scheme

[0197] 74 36 52 21 67 79 15 18 85 70 29 19 9 91 49

[0198] Table 5 decoding results of remainder random selection model ​

[0199]

Claims

1. A method for full-coverage autonomous inspection and scheduling of power transmission lines using unmanned aerial vehicles (UAVs) based on the frog-leaping algorithm, characterized in that, Includes the following steps: Step 1: Obtain the numbers of the transmission line towers to be inspected within the inspection area, as well as the workload of different types of tower inspection tasks; Step 2: Determine the drone number used for pole inspection, the number of drones available for each type of inspection task and their specific drone numbers, the inspection speed of each drone and the corresponding degree of inspection quality defects; Step 3: Establish an optimal scheduling model for full-coverage autonomous inspection scheduling of UAVs; Step 3.1: Determine the optimization objective of the optimal scheduling model; Step 3.1.1: The first objective function of constructing the optimal scheduling model is to find the minimum value of the maximum completion time for the UAV to complete the full-coverage autonomous inspection; Step 3.1.2: The second objective function of the optimization scheduling model is to find the minimum value of the average value of the defective quality of UAV inspections; Step 3.2: Establish the constraints for the optimal scheduling model; Step 4: Develop a two-layer coding mechanism that includes information on task tower sorting and UAV inspection speed selection; Step 5: Based on the characteristics of the aforementioned optimal scheduling model, the hybrid frog-jumping algorithm is improved, and an external archive maintenance method based on iterative repetition rate is proposed. Step 6: Use the improved hybrid frog-jumping algorithm obtained in Step 5 to solve the optimal scheduling model in Step 3, and obtain the optimal inspection scheduling scheme.

2. The method for full-coverage autonomous inspection and scheduling of power transmission lines by unmanned aerial vehicles according to claim 1, characterized in that, In step 1, the different types of tower inspection tasks include tower autonomous fine-tuning inspection tasks, conductor and ground wire autonomous inspection tasks, and channel autonomous inspection tasks.

3. The method for full-coverage autonomous inspection and scheduling of power transmission lines using unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, In step 2, n represents the number of poles to be inspected within the area, x represents the number of the poles to be inspected, and U represents the UAV ensemble. k This represents the set of drones used for the k-th type of inspection task. k represents the type of inspection, i.e. the inspection stage. k=1 represents the autonomous and refined inspection task of the tower, k=2 represents the autonomous inspection task of the conductor and ground wire, and k=3 represents the autonomous inspection task of the channel. u k This represents the number of drones used for the k-th type of inspection task; Represents the j-th type of inspection task of the k-th category. k The serial number of the drone Indicates the i-th x The k-th type of inspection task for each tower, v jk Indicates drone Inspection speed, Indicates inspection task The inspection workload, V represents the inspection speed. jk The set of drone mission inspection speeds Indicates drone For the i-th x The time required for each tower to complete the k-th type of inspection task. Indicates inspection task The start time of the inspection. Indicates inspection task Inspection completion time, F max Indicates the maximum completion time. U-shaped drone jk With speed The degree of poor quality of the patrol inspection.

4. The method for full-coverage autonomous inspection and scheduling of power transmission lines by unmanned aerial vehicles according to claim 3, characterized in that, Settings: All drones in the drone collection U can be used for inspection operations. A single drone can only perform one type of inspection task on one pole at a time and must not be interrupted. The number of drones u used for the k-th type of inspection task k Not less than 1; The workload of the path flight from the take-off point or the end point of the previous inspection task to the transmission line inspection task point is included in the workload of the current inspection task. i x =1,2,...,n,i′ x =1,2,...,n (1) j k =1,2,...,u k (2) k = 1, 2, or 3 (3) Equations (6) and (7) represent the binary decision-making of the task tower inspection; Equations (8) and (9) represent the priority relationship between towers in the same inspection stage; Equations (10) and (11) represent the priority relationship between different inspection stages of the same tower; Equation (12) defines the maximum completion time as equal to the time of completion of the last inspection; Equation (13) ensures that the start time of the first inspection stage is greater than or equal to zero; Equation (14) ensures that the UAV inspects each tower at the selected inspection speed in any inspection stage; Equation (15) calculates the completion time of any inspection stage; Equation (16) sets constraints to ensure that the same tower must complete the inspection task of the current inspection stage before proceeding to the next inspection stage; Equations (17) and (18) ensure the constraint relationship between different towers in the same inspection stage; Equations (19) and (20) ensure the sequential constraint relationship of different inspection stages of the same tower.

5. The method for full-coverage autonomous inspection and scheduling of power transmission lines by unmanned aerial vehicles according to claim 4, characterized in that, In step 3.1.1, the minimum maximum completion time for the full-coverage autonomous inspection is: In the formula, f1 represents the first objective function; Indicates the i-th x The inspection task of a tower in the k-th inspection phase The completion time.

6. The method for full-coverage autonomous inspection and scheduling of power transmission lines by unmanned aerial vehicles according to claim 5, characterized in that, In step 3.1.2, calculate the average value that minimizes the degree of defective inspection quality: In the formula, f2 represents the second objective function.

7. The method for full-coverage autonomous inspection and scheduling of power transmission lines by unmanned aerial vehicles according to claim 6, characterized in that, The scheduling scheme of the inspection scheduling method adopts a two-layer coding mechanism; The scheduling scheme uses a two-layer coding system. The first layer is the tower inspection sequence coding: X1X2X3…X n ,in i x =1,2,…n indicates that the inspection sequence number is i x The number of poles to be inspected is n, where n is the number of poles to be inspected. The second layer of the scheduling scheme is a constraint code, which represents the selection information of the drone inspection speed for different inspection stages of each tower.

8. The method for full-coverage autonomous inspection and scheduling of power transmission lines by unmanned aerial vehicles according to claim 7, characterized in that, In step 5, the improved hybrid frog jumping algorithm includes the following steps: (a) Initialize the population and external files; Initialize the frog population P = {O1, O2, ..., O} N }, O p p = 1, 2, ..., N represents the p-th frog, where each frog in the frog population corresponds to a feasible solution of the optimal scheduling model. O(n) p =[O pα O pβ] T Among them O pα This represents the first objective function value for the p-th frog, O. pβ This represents the value of the second objective function for the p-th frog. The frog population initialization parameters use a random number generation method to ensure the dispersion of individuals in the initial frog population; external files are initialized. (ii) Constructing a frog individual relationship pairing matrix; For each individual frog O p Calculate the variant index value L p Individuals pair up in pairs, and the p1th frog in the population... With the p2th frog During pairing, the label value is set based on the comparison result of the variant index values. remember This represents the p1th frog in the population. With the p2th frog The paired label values; if remember like remember The label values ​​of the pairwise pairings between individuals are used as elements in the frog individual relationship pairing matrix to construct the frog individual relationship pairing matrix; (III) Construct and train the radius neighbor classifier (RNC); The frog individual relationship pairing matrix is ​​used as the training data, and the label values ​​in the frog individual relationship pairing matrix are used as the training data labels. A radius neighbor classifier (RNC) is constructed and trained. The parameter value of the radius threshold (Radius) is determined by hyperparameter search. (iv) Divide the population into multiple levels; The population is stratified according to the dominance relationship in the population. The first layer consists of all non-dominated solutions, the second layer consists of solutions dominated by the first layer, and so on, so that the frog population is divided into multiple layers. Using the fast nondominated solution sorting algorithm, for the p-th frog individual O p Assigning layering parameter R p Frog individuals with the same stratification parameter value are grouped into the same stratum, p = 1, 2, ..., N. The maximum number of strata is Lmax, where Lmax represents the number of strata in the frog population. If R p =l, then O p They were assigned to the first level; (v) Calculate the congestion level; For the stratified frog population, calculate the O of the p-th frog individual. p Crowding value C relative to other individuals in the population p It is used to assess the surrounding density of individual frogs in a population; For frog individuals with the same stratification parameter value in the population, the frog individual with a larger crowding value is better; Let Cmax be the crowding value of the frog individual with the highest crowding value in the l-th layer of the frog population. l The crowding value of the frog with the lowest crowding value is denoted as Cmin. l l = 1, 2, ..., Lmax, where Lmax represents the number of frog population levels; (vi) Retain the best individual in the hierarchy; Sort the frogs in the l-th layer of the frog population from largest to smallest based on their crowding density. Then, extract the top n frogs from the resulting sequence. l n frog individuals are grouped into an elite set E and placed into the next generation population Q. l This refers to the number of elite frog individuals in the l-th stratum of the frog population. And n l ∈Z (vii) Constructing meme groups; Frog individuals in the next generation population Q are sorted according to their crowding scores and then evenly distributed to m. Q m different meme groups Q Indicates the number of memes; (viii) Meme group evolution and individual renewal; Mutation and crossover operations are performed on individual frogs in each meme group to generate a new solution O. new ; for the newly generated solution O new and the old solutions O in the existing meme set org Pairing is performed, and the Radius Neighbor Classifier (RNC) trained in step (iii) is used to predict the labels of the relationships between individual frogs. If the predicted label value is 1 or 0, it indicates that a new solution O is found. new Better than or equal to the old solution O org Then use the new solution O new Replace old solution O org If the predicted label value is -1, it indicates that the new solution is O. new Inferior to the old solution O org Then a random solution is generated to replace the old solution O. org ; (ix) External archive maintenance; Add all individuals of the evolved meme group to the external file; if the external file exceeds the capacity limit, perform a maintenance operation of filtering and deleting. (x) Determine whether the algorithm termination condition is met; Randomly select solutions of proportion Fr from the external archives to form the next generation of new population P', where Fr represents the proportion of solutions in the external archives. Determine whether the algorithm termination condition is met. If the termination condition is met, stop the search and return the non-dominated solution; otherwise, proceed to step (iv).

9. The method for full-coverage autonomous inspection and scheduling of power transmission lines by unmanned aerial vehicles according to claim 8, characterized in that, The proportion of solutions extracted from external archives Fr∈(0.5,1)。 10. The method for full-coverage autonomous inspection and scheduling of power transmission lines by unmanned aerial vehicles according to claim 9, characterized in that, Based on the iteration repetition rate, external archives are maintained, specifically including: For individual frog O that enters the external archive p For non-dominated solutions, set the corresponding counters for p = 1, 2, ... N. p Counter p O represents the p-th frog individual. p The corresponding counter is used to record the number of times the same non-dominated solution appears in the iteration. When the capacity of the external archive reaches a fixed proportion k of the upper limit, the counter is used. f At that time, k f This represents the proportion corresponding to the maximum capacity of external files; begin calculating the non-dominated solution O. p Iteration repetition rate IR p And according to the corresponding elimination probability EP p Implement elimination; IR p =Counter p / Iter now EP0 = 0, EP1 < EP2 < 1 In the formula Iter now Indicates the current iteration number; IR p O represents the p-th frog individual. p The iteration repetition rate of non-dominated solutions; EP p O represents the p-th frog individual. p The elimination probability of a non-dominated solution; EP0 represents the elimination probability of a small iteration repetition rate, EP1 represents the elimination probability of a medium-to-high iteration repetition rate, and EP2 represents the elimination probability of a large iteration repetition rate.

Images