A multi-target path planning method for a power station inspection robot

Abstract

The application discloses a multi-target path planning method of a power station inspection robot. Path planning is a key link for operation of the power station inspection robot system, and the performance of the path planning is closely related to equipment inspection efficiency and energy consumption ratio. Current mainstream algorithms generally have defects such as slow convergence speed, poor safety and low planning efficiency. Compared with the traditional method, the application significantly improves the path planning efficiency and path quality by simultaneously optimizing multiple targets, and enhances the stability and safety of the path.

Description

Technical Field

[0001] This invention relates to the field of robot path planning, and more specifically to a multi-objective path planning method for a power plant inspection robot. Background Technology

[0002] As a transit point for power transmission and distribution, the safe and stable operation of power plants is crucial to ensuring safe production. With the expansion of substation scale and technological advancements, robots are gradually replacing manual labor in inspections. Path planning, one of the most important technologies in the field of inspection robots, helps robots avoid obstacles during their journey and quickly reach their destination. However, path planning still relies heavily on manual intervention in many cases. This semi-manual planning method is susceptible to subjective factors, leading to a waste of manpower, resources, and time. Therefore, in today's era of industry-wide intelligence, researching intelligent path planning algorithms and applying them to the field of mobile robots is of great significance. Summary of the Invention

[0003] Purpose of the invention: In order to overcome the limitations of traditional path planning algorithms, this invention provides a multi-objective path planning method for power plant inspection robots, which significantly improves path planning efficiency and path quality, and enhances path stability and safety.

[0004] Technical solution: The multi-objective path planning method for the power plant inspection robot of the present invention includes:

[0005] (1) Population initialization: determine the population size, select the encoding method, and generate the initial population;

[0006] (2) Perform selection, mutation, and crossover operations on the initial population;

[0007] (3) Generate offspring using the non-dominated sorting and elite strategy in the non-dominated sorting genetic algorithm;

[0008] (4) The population individuals are sorted and selected using a linear ranking-based selection strategy to generate new offspring;

[0009] (5) The population is selected, mutated and crossovered again through a phased transformation strategy based on a probability roadmap;

[0010] (6) If the set number of iterations is reached, proceed to step (7); otherwise, proceed to step (4).

[0011] (7) Use a smoothing algorithm to smooth the path;

[0012] (8) The process ends when the destination is reached.

[0013] Furthermore, including:

[0014] In step (3), the non-dominated ranking aims to hierarchically rank the individuals in the population. One solution is non-dominated by another, meaning that it is superior to the latter in at least one objective and not inferior to the latter in other objectives. Stratification occurs when a large number of individuals satisfy both dominance and non-dominance. This process can be achieved by stratifying all individuals layer by layer, ensuring that all individuals are assigned to a suitable frontier rank.

[0015] Crowding distance measures the degree of crowding of an individual at its forefront. Individuals with lower crowding have a greater probability of selection, thus promoting the diversity of solutions in the population. Calculating crowding distance requires first sorting each target, and then assigning the maximum crowding value to the individuals at the very front and rear of the target. For each target... Individuals in Its congestion The calculation is expressed as:

[0016] (1),

[0017] in, and solutions respectively In the target Adjacent solutions on the target. The crowding distance calculation process can be performed on each target to ensure the crowding degree of each solution.

[0018] The non-dominated sorting genetic algorithm employs an elitist strategy to ensure that the best individuals are retained in each generation. This strategy involves merging parent and offspring individuals and selecting a new population based on non-dominated sorting and crowding distance. This strategy avoids the loss of excellent individuals due to random selection, improving the algorithm's convergence and stability.

[0019] Furthermore, including:

[0020] In step (4), the linear ranking-based selection strategy abandons the random selection of individuals and instead determines which individuals can reproduce based on their fitness ranking in the population. Individuals in the population are sorted by fitness from smallest to largest. In the non-dominated sorting genetic algorithm, the ranking is first sorted in descending order, and then in ascending order based on crowding distance. Higher rankings correspond to lower fitness and higher selection probabilities, and a linear function is used to allocate the selection probability for each individual. Let... Individuals in a population, among which Ranked highest Lowest ranking, individual The probability of selection is

[0021] (2),

[0022] in, and All are constants, usually , .if Then at this time It is a random selection, and the population selection pressure is minimized; when , At this point, the selection pressure on the population is greatest. This can be gradually adjusted by adding an exponential term as the generations progress. and The value effectively increases the selection pressure on the parent generation.

[0023] To maintain population diversity and enhance the selection of superior individuals, an activation function is used to optimize the calculation of the selection probability. First, a selection probability score needs to be defined. for:

[0024] (3),

[0025] Applying the ReLU function, we get:

[0026] (4),

[0027] After normalizing the score, we can obtain:

[0028] (5),

[0029] in, It is the sum of the scores of all individuals. Individuals after applying activation functions The score. From the above derivation process, the final selection probability can be obtained as:

[0030] (6),

[0031] Here, the present invention performs additional scaling on the normalized score to ensure that the final probability value falls within the expected range.

[0032] Furthermore, including:

[0033] In step (5), the population is selected, mutated and crossovered again through a phased transformation strategy based on a probability roadmap.

[0034] In the selection phase, connection information from the probabilistic path graph is used to select parents. Combining this with path information from the probabilistic path graph, individuals connected to the current node are prioritized as offspring, or individuals connected to the target node are prioritized as parents. In the mutation phase, the probabilistic path graph algorithm is used to maintain population diversity. In the initial mutation phase, the target of the mutation operation needs to be determined; the target can be a node or a connection between two nodes. In the crossover phase, new feasible paths are generated by exchanging portions of two selected paths within the population.

[0035] Furthermore, including:

[0036] In step (7), the improved Chaikin algorithm is used to smooth the path.

[0037] The standard Chaikin algorithm is a direct corner-cutting technique used to transform control polygons into smooth curves. Its core idea is to... (The sentence is incomplete and requires more context to translate accurately.) Take two new control points and , through connection and This replaces the original path segment, thus removing sharp corners. In the standard Chaikin algorithm, and The position is usually determined by a fixed interpolation factor. The core improvement of the method proposed in this invention lies in the introduction of an adaptive smoothing factor based on the angle between continuous path segments. .

[0038] In a given set of waypoints In the middle, for each point on the path (in ), calculate the connecting line segments and The angle between This invention is based on a preset angle threshold. Perform conditional judgments to determine the smoothing factor. .like Then let Set to 0.1 (low smoothness for acute angles), otherwise set to The value is 0.3 (obtuse angles use high smoothness). For each pair of consecutive points... and Two new control points are generated. and For point and , and Defined as:

[0039] (7),

[0040] (8),

[0041] Then the original point set is replaced with the newly generated point set to obtain a new point set. Repeat the above steps for a new set of points until the preset number of iterations is reached, and finally obtain the smoothed path.

[0042] Furthermore, including:

[0043] In step (8), if the target has been reached, the planning process ends; otherwise, the path needs to be replanned until the target is reached. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the path planning algorithm proposed in this invention;

[0045] Figure 2 This is a schematic diagram of the probability roadmap transformation during the selection phase;

[0046] Figure 3 This is a schematic diagram of the probability roadmap transformation during the mutation stage;

[0047] Figure 4 This is a schematic diagram of the probability roadmap transformation during the crossover phase. Detailed Implementation

[0048] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The specific description is as follows:

[0049] like Figure 1 As shown in the diagram, the overall algorithm flowchart of this invention is as follows:

[0050] Step 1: Population initialization, determining the population size, selecting the encoding method, and generating the initial population.

[0051] Step 2: Perform selection, mutation, and crossover operations on the initial population.

[0052] Step 3: Generate offspring using the non-dominated sorting and elite strategy in the non-dominated sorting genetic algorithm.

[0053] The calculation process in step 3 is as follows:

[0054] Non-dominated ranking aims to hierarchically rank individuals in a population. One solution is non-dominated by another, meaning it is superior to the latter in at least one objective and not inferior to it in others. Stratification occurs when a large number of individuals satisfy both dominance and non-dominance. This process can be achieved by stratifying all individuals layer by layer, ensuring that all individuals are assigned to an appropriate frontier rank.

[0055] Crowding distance measures the degree of crowding of an individual at its forefront. Individuals with lower crowding have a greater probability of selection, thus promoting the diversity of solutions in the population. Calculating crowding distance requires first sorting each target, and then assigning the maximum crowding value to the individuals at the very front and rear of the target. For each target... Individuals in Its congestion The calculation is expressed as:

[0056] (1),

[0057] in, and solutions respectively In the target Adjacent solutions on the target. The crowding distance calculation process can be performed on each target to ensure the crowding degree of each solution.

[0058] The non-dominated sorting genetic algorithm employs an elitist strategy to ensure that the best individuals are retained in each generation. This strategy involves merging parent and offspring individuals and selecting a new population based on non-dominated sorting and crowding distance. This strategy avoids the loss of excellent individuals due to random selection, improving the algorithm's convergence and stability.

[0059] Step 4: Sort and select individuals in the population using a linear ranking-based selection strategy to generate new offspring.

[0060] The calculation process in step 4 is as follows:

[0061] The linear ranking-based selection strategy abandons random selection and instead determines which individuals can reproduce based on their fitness ranking within the population. Individuals are sorted in ascending order of fitness. In the non-dominated sorting genetic algorithm, the ranking is first sorted in descending order of fitness, then in ascending order of crowding distance. Higher rankings correspond to lower fitness and higher selection probabilities. A linear function is then used to allocate the selection probability for each individual. Let... Individuals in a population, among which Ranked highest Lowest ranking, individual The probability of selection is:

[0062] (2),

[0063] in, and All are constants, usually , .if Then at this time It is a random selection, and the population selection pressure is minimized; when , At this point, the selection pressure on the population is greatest. This can be gradually adjusted by adding an exponential term as the generations progress. and The value effectively increases the selection pressure on the parent generation.

[0064] To maintain population diversity and enhance the selection of superior individuals, the ReLU activation function is used to optimize the calculation of selection probability. First, the selection probability score needs to be defined. for:

[0065] (3),

[0066] Next, applying the ReLU function, we get:

[0067] (4),

[0068] After normalizing the score, we can obtain:

[0069] (5),

[0070] in, It is the sum of the scores of all individuals. Individuals after applying activation functions The score. From the above derivation process, the final selection probability can be obtained as:

[0071] (6),

[0072] Here, the present invention performs additional scaling on the normalized score to ensure that the final probability value falls within the expected range.

[0073] Step 5: The population is selected, mutated, and crossovered again through a phased transformation strategy based on a probability roadmap.

[0074] The calculation process for step 5 is as follows:

[0075] The population is further selected, mutated, and crossovered using a phased transformation strategy based on a probabilistic roadmap. In the selection phase, connection information from the probabilistic roadmap is used to select parents. Combined with path information in the probabilistic roadmap, individuals connected to the current node are prioritized as offspring, or individuals connected to the target node are prioritized as parents. In the mutation phase, the probabilistic roadmap algorithm maintains population diversity. At the initial stage of mutation, the target of the mutation operation needs to be determined; the target can be a node or a connection between two nodes. In the crossover phase, new feasible paths are generated by exchanging portions of two selected paths within the population. See attached details. Figure 2 Appendix Figure 3 and appendix Figure 4 As shown.

[0076] Step 6: If the set number of iterations is reached, proceed to step (7); otherwise, proceed to step (4).

[0077] Step 7: Smooth the path using an improved Chaikin algorithm.

[0078] The calculation process for step 7 is as follows:

[0079] The standard Chaikin algorithm is a direct corner-cutting technique used to transform control polygons into smooth curves. Its core idea is to... (The sentence is incomplete and requires more context to translate accurately.) Take two new control points and Then through connection and This replaces the original path segment, thus removing sharp corners. In the standard Chaikin algorithm, and The position is usually determined by a fixed interpolation factor. The core improvement of the method proposed in this invention lies in the introduction of an adaptive smoothing factor based on the angle between continuous path segments. .

[0080] In a given set of waypoints In the middle, for each point on the path (in ), calculate the connecting line segments and The angle between This invention is based on a preset angle threshold. Perform conditional judgments to determine the smoothing factor. .like Then let Set to 0.1 (low smoothness for acute angles), otherwise set to The value is 0.3 (obtuse angles use high smoothness). For each pair of consecutive points... and Two new control points are generated. and For point and , and Defined as:

[0081] (7),

[0082] (8),

[0083] Then the original point set is replaced with the newly generated point set to obtain a new point set. Repeat the above steps for a new set of points until the preset number of iterations is reached, and finally obtain the smoothed path.

[0084] Step 8: If the destination has been reached, the planning process ends; otherwise, the route needs to be replanned until the destination is reached.

[0085] This invention employs a multi-objective path planning method based on probabilistic route graph-nondominated sorting genetic algorithm, which significantly improves path planning efficiency and path quality compared to traditional methods, and enhances the stability and security of the power plant inspection system.

[0086] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A multi-objective path planning method for a power plant inspection robot, characterized in that, Includes the following steps: (1) Population initialization: determine the population size, select the encoding method, and generate the initial population; (2) Perform selection, mutation, and crossover operations on the initial population; (3) Generate offspring using the non-dominated sorting and elite strategy in the non-dominated sorting genetic algorithm; (4) The population individuals are sorted and selected using a linear ranking-based selection strategy to generate new offspring; (5) The population is selected, mutated and crossovered again through a phased transformation strategy based on a probability roadmap; (6) If the set number of iterations is reached, proceed to step (7); otherwise, proceed to step (4). (7) Use a smoothing algorithm to smooth the path; (8) The process ends when the destination is reached.

2. The method according to claim 1, characterized in that, In step (3), for the target Individuals in Its congestion The calculation is expressed as: （1）， in, This indicates the number of objective functions in a multi-objective optimization problem. and solutions respectively In the target Adjacent solutions on and solutions respectively In the target The function values ​​corresponding to adjacent solutions on the [theory / method].

3. The method according to claim 1, characterized in that, In step (4), there is a set Individuals in a population, among which Ranked highest Lowest ranking, individual The probability of selection is: （2）， in, and All are constants, usually , ,if Then at this time It is a random selection, and the population selection pressure is minimized; when , At this point, the selection pressure on the population is greatest, and by adding an exponential term, the selection can be gradually adjusted as the generations progress. and The value, First, define the selection probability score. for: （3）， Then, applying the function, we get: （4）， After normalizing the score, we can obtain: （5）， in, It is the sum of the scores of all individuals. Individuals after applying the function Based on the above derivation process, the final selection probability can be obtained as: （6）， The normalized scores were further scaled, and the final probability values ​​were within the expected range.

4. The method according to claim 1, characterized in that, In step (5), the population is selected, mutated, and crossovered again through a phased transformation strategy based on a probability roadmap; During the selection phase, the connection information in the probability roadmap is used to select the parent generation. Combined with the path information in the probability roadmap, individuals connected to the current node are given priority as offspring, or individuals connected to the target node are given priority as parents.

5. The method according to claim 1, characterized in that, In step (7), the improved Chaikin algorithm is used to smooth the path; In a given set of waypoints In the middle, for each point on the path ,in Calculate the connecting line segments and The angle between Based on a preset angle threshold Perform conditional judgments to determine the smoothing factor. ,like Then let Set to 0.1, otherwise set to The value is 0.3 for each pair of consecutive points. and Two new control points are generated. and For point and , and Defined as: （7）， （8）， Then the original point set is replaced with the newly generated point set to obtain a new point set. Repeat the above steps for a new set of points until the preset number of iterations is reached, and finally obtain the smoothed path.

6. The method according to claim 1, characterized in that, In step (8), if the target has been reached, the planning process ends; otherwise, the path needs to be replanned to eventually reach the target.

Images