A substation inspection robot obstacle crossing action sequence optimization method based on NSGA-II and fuzzy neural network

Abstract

The application is a kind of substation inspection robot obstacle crossing action sequence optimization method based on NSGA-Ⅱ and fuzzy neural network, first, the size parameters of the substation inspection robot are determined and the substation inspection robot obstacle crossing dynamics model is established; secondly, based on NAGA-Ⅱ algorithm, the substation inspection robot obstacle crossing multi-objective optimization model is established, taking the smallest possible drive torque peak and the smallest possible battery energy consumption in the obstacle crossing process as the multi-objective optimization principle, and the multi-objective optimization Pareto frontier under different obstacle heights H is obtained by solving; finally, the substation inspection robot obstacle crossing dynamic fuzzy neural network is established, and the Pareto frontier is used as the learning sample, so that the optimized obstacle crossing action sequence under different obstacle heights is obtained by training. The application has important significance in increasing the endurance time of the substation inspection robot, enhancing the system operation reliability, improving the system intelligence degree, increasing the work efficiency, expanding the application field and the like.

Description

Technical Field

[0001] This invention relates to a method for optimizing the obstacle-crossing action sequence of a substation inspection robot, specifically a method based on NSGA-II and fuzzy neural networks, belonging to the field of service robot technology. Background Technology

[0002] Currently, substations require regular inspections and maintenance to ensure the reliability of power distribution. Frequent inspections can detect equipment problems early, minimizing potential losses. Currently, the vast majority of inspections are done manually, requiring designated personnel to conduct inspections at regular intervals within the substation. This method is resource-intensive and prone to human error, which can cause damage to the substation or even lead to safety accidents. With the development of service robot technology, some ground robots are beginning to be used for autonomous substation inspections.

[0003] However, the current technology level of inspection robots varies greatly. Although most robots employ multi-sensor data fusion and intelligent control theory to achieve a high degree of intelligence, they often encounter obstacles while following a set route, requiring them to overcome these obstacles. On the one hand, substation inspection robots are relatively lacking in system endurance and motor reliability. This often results in short robot endurance and motor damage due to prolonged high-load operation. On the other hand, whether it's line-of-sight operation, remote operation, or autonomous obstacle crossing, the action sequences of substation inspection robots are often based on expert experience, and the robots cannot intelligently infer the action sequences for obstacle crossing.

[0004] Therefore, designing an optimization method for the obstacle-crossing action sequence of a substation inspection robot system, which addresses the energy consumption and operational reliability of the system, is of great significance in increasing the robot's endurance, enhancing system reliability, improving system intelligence, increasing work efficiency, and expanding application areas. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method for optimizing the obstacle-crossing action sequence of a substation inspection robot based on NSGA-II and fuzzy neural network, so that the substation inspection robot and the like can complete obstacle crossing in the best state.

[0006] An optimization method for obstacle-crossing motion sequences of a substation inspection robot based on NSGA-II and fuzzy neural networks includes the following steps:

[0007] S1: determining the substation inspection robot size parameters, step obstacle size parameters, including: vehicle length, wheel radius, vehicle mass, mass center position, front and rear swing arm length, swing arm wheel radius, swing arm mass, obstacle height;

[0008] S2: establishing a substation inspection robot obstacle dynamics model, the input variables are the substation inspection robot size parameters, obstacle height, and obstacle parameters. The output results are the driving torque peak value in the obstacle process and the battery energy consumption in the obstacle process;

[0009] S3: based on the NSGA-II algorithm, an obstacle multi-objective optimization model of the substation inspection robot is established. The optimization model independent variables are: two obstacle parameters θ0 and θ2. The optimization model optimization goals are: the driving torque peak value in the obstacle process is as small as possible, and the battery energy consumption in the obstacle process is as small as possible;

[0010] S4: the NSGA-II algorithm is used to solve the multi-objective optimization model, and thus the multi-objective optimization Pareto front under different obstacle heights H is obtained;

[0011] S5: a fuzzy neural network of the substation inspection robot obstacle dynamics is established, the fuzzy neural network input is the obstacle height, the battery energy consumption in the obstacle process, and the driving torque peak value, the fuzzy neural network output is the obstacle parameter under the corresponding input condition, and the fuzzy neural network learning sample is the Pareto front.

[0012] The obstacle dynamics model in step S2 is for the process from when the substation inspection robot contacts the obstacle with the front swing arm to when the mass center completely passes the obstacle angle line. The obstacle parameter is the angle between the swing arm and the horizontal plane during the two times of lifting the swing arm in the robot obstacle process.

[0013] The substation inspection robot obstacle multi-objective optimization model in step S3 is specifically:

[0014]

[0015] Wherein, W is the battery energy consumption in the robot obstacle process, F is the driving torque peak value in the obstacle process, θ0 and θ2 are the substation inspection robot obstacle parameters, F1 and F2 are the driving forces provided by the robot on the ground and the obstacle in the obstacle process, X is the distance from the robot contact point to the obstacle, α is the angle between the robot body axis and the ground, and i is the obstacle stage.

[0016] The multi-objective optimization model has two optimization goals. The first optimization goal is to make the battery energy consumption in the obstacle process as small as possible, that is, W takes the minimum value as much as possible. The second optimization goal is to make the driving torque peak value in the obstacle process as small as possible, that is, F takes the minimum value as much as possible. The optimization goals are:

[0017]

[0018] The solving of the multi-objective optimization model in the step S4 uses the NSGA-II algorithm, and the sequence and content are divided into 9 sub-processes:

[0019] S41: Set the specific obstacle height H, determine the value range of the individual according to the value range of the obstacle parameter, and then initialize the population;

[0020] S42: Perform non-dominated sorting and crowding degree calculation on the initialized population, and each individual has its own rank and crowding degree.

[0021] S43: Set the iteration number, and stop iteration if the iteration number is exceeded;

[0022] S44: Compare the ranks of the individuals two by two, select the individual with a smaller rank, and if the ranks are the same, select the individual with a larger crowding degree,

[0023] S45: Obtain a new population by crossover, mutation and population merging;

[0024] S46: Calculate the current non-dominated sorting and crowding degree again;

[0025] S47: Update the population, and generate a new population through the elite strategy competition;

[0026] S48: Determine whether the maximum iteration number is reached, and if not, return to the fourth process to continue the steps S44-S48;

[0027] S49: The algorithm ends, and the Pareto front under each obstacle height H is obtained.

[0028] The sequence and content of establishing the obstacle fuzzy neural network of the substation inspection robot in the step S5 are divided into the following 3 sub-processes:

[0029] S51: Set the network input nodes, hidden nodes and output nodes. The input nodes have 3, which are obstacle height H, battery energy consumption W in the robot obstacle process, and driving torque peak value F in the obstacle process. The output nodes have 2, which are the first arm lifting angle θ0 and the second arm lifting angle θ2;

[0030] S52: After normalizing the Pareto front under each obstacle height H in the step S4, use it as a sample of fuzzy rules in the fuzzy neural network, and input the sample into the fuzzy neural network for training.

[0031] S53: training the fuzzy neural network to obtain the obstacle parameters under any obstacle height H, battery energy consumption W, and maximum driving force F, i.e., the first swing arm lifting angle θ0 and the second swing arm lifting angle θ2, so as to obtain the optimized obstacle action sequence of the substation inspection robot under different obstacle heights.

[0032] Compared with the prior art, the present application has the following advantages:

[0033] (1) The present application can effectively reduce the maximum driving force of the driving motor during the obstacle crossing process of the substation inspection robot, effectively reduce the working pressure of the driving motor of the substation inspection robot, and thus improve the reliability of the driving motor and the reliability of the entire substation inspection robot;

[0034] (2) The present application can effectively reduce the battery energy consumption during the obstacle crossing process of the substation inspection robot, can significantly reduce the energy consumption during the obstacle crossing of the substation inspection robot, and effectively improve the endurance time of the substation inspection robot;

[0035] (3) The present application can perform multi-objective optimization on the driving torque peak value during the obstacle crossing process of the substation inspection robot and the battery energy consumption during the obstacle crossing process, greatly reduce the search space, improve the search efficiency, and quickly obtain the best parameter combination of the substation inspection robot obstacle crossing, i.e., the Pareto frontier, while meeting the conditions that the driving torque peak value during the obstacle crossing process and the battery energy consumption during the obstacle crossing process are as small as possible;

[0036] (4) The present application uses the Pareto frontier as the learning sample of the fuzzy neural network, can specifically train the fuzzy neural network, obtain the obstacle action sequence under the required parameters, fully utilize the advantages of expert knowledge and reasoning ability, effectively improve the intelligent level of the substation inspection robot, and has important significance in increasing the endurance time of the substation inspection robot, enhancing the system operation reliability, improving the system intelligent degree, increasing the work efficiency, expanding the application field, etc. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 It is a flowchart of a substation inspection robot obstacle action sequence optimization method based on NSGA-II and fuzzy neural network;

[0038] Figure 2 It is a substation inspection robot obstacle flowchart;

[0039] Figure 3 It is a force analysis diagram of the substation inspection robot obstacle dynamics model;

[0040] Figure 4 It is a substation inspection robot obstacle dynamics solution result diagram;

[0041] Figure 5 Flow chart for solving multi-objective optimization model by using NSGA-II algorithm;

[0042] Figure 6 Schematic diagram of Pareto front obtained by using NSGA-II algorithm for optimization;

[0043] Figure 7 Flow chart of fuzzy neural network algorithm for obstacle crossing of substation inspection robot; DETAILED DESCRIPTION

[0044] The application will be further described in detail below in combination with the drawings and specific embodiments:

[0045] As shown in the drawings, a substation inspection robot obstacle crossing action sequence optimization method based on NSGA-II and fuzzy neural network comprises the following steps: Figure 1 S1: Determine the size parameters of the substation inspection robot and the size parameters of the step obstacle, including: vehicle body length, wheel radius, vehicle body mass, center of mass position, front and rear swing arm length, swing arm wheel radius, swing arm mass, obstacle height;

[0046] S2: Establish the obstacle crossing dynamics model of the substation inspection robot, and input the size parameters of the substation inspection robot, the obstacle height, and the obstacle crossing parameters. The output result is the driving torque peak value in the obstacle crossing process and the battery energy consumption in the obstacle crossing process;

[0047] S3: Based on the NAGA-II algorithm, establish a multi-objective optimization model for the obstacle crossing of the substation inspection robot. The independent variables of the optimization model are: two obstacle crossing parameters θ0 and θ2. The optimization objectives of the optimization model are: the driving torque peak value in the obstacle crossing process is as small as possible, and the battery energy consumption in the obstacle crossing process is as small as possible;

[0048] S4: Use the NSGA-II algorithm to solve the multi-objective optimization model, and thus obtain the multi-objective optimized Pareto front under different obstacle heights H;

[0049] S5: Establish a fuzzy neural network for the obstacle crossing of the substation inspection robot. The input of the fuzzy neural network is the obstacle height, the battery energy consumption in the obstacle crossing process, and the driving torque peak value. The output of the fuzzy neural network is the obstacle crossing parameter under the corresponding input condition. The learning sample of the fuzzy neural network is the Pareto front.

[0050] The obstacle crossing dynamics model in the step S2 is for the process from when the front swing arm of the substation inspection robot contacts the obstacle to when the center of mass completely crosses the obstacle angle line. As shown in the drawings, wherein the obstacle crossing parameter is the angle between the swing arm and the ground when the swing arm is lifted twice during the robot obstacle crossing process.

[0051] Figure 2

[0052] ​​The obstacle-crossing action sequence of the obstacle-crossing dynamics model is as shown in the following table. Figure 2 The obstacle-crossing action sequence of the obstacle-crossing dynamics model is as shown in the following table.

[0053] S21: The swing arm of the substation inspection robot is raised to a certain angle, and the ground reaches a certain angle, and advances to touch the obstacle. The angle is the first obstacle-crossing parameter θ0;

[0054] S22: The driving motor drives the substation inspection robot to advance, and the angle between the vehicle body and the ground starts to increase until the vehicle body part just reaches the angle line of the obstacle;

[0055] S23: The swing arm of the substation inspection robot is adjusted again, and the angle with the ground becomes another value θ2;

[0056] S24: The driving motor drives the substation inspection robot to advance, and the angle between the vehicle body and the ground increases to the angle line of the obstacle;

[0057] S25: The fifth process is that the robot rotates around the angle line of the obstacle, and the obstacle-crossing is successful.

[0058] The force analysis of the obstacle-crossing dynamics model is as shown in the following table. Figure 3 The parameters of the dynamics model are determined by the parameter values in step S1. The solution of the dynamics model is that, for the above five sub-processes: S21, S22, S23, S2, S25, as the angle α between the robot vehicle body and the ground increases, the driving forces F1 and F2 provided by the robot on the ground part and the obstacle part during the obstacle-crossing process, and the required coefficient μ. Set the height of a certain obstacle and the obstacle-crossing parameters, and the dynamics solution is as shown in the following table. Figure 4

[0059] The whole obstacle-crossing process of the substation inspection robot can be mathematically modeled by the step S2, thereby providing theoretical support for the subsequent multi-objective optimization model.

[0060] The multi-objective optimization model of the substation inspection robot obstacle-crossing in step S3 is as follows:

[0061]

[0062] Wherein, W is the battery energy consumption during the robot obstacle-crossing process, F is the driving torque peak value during the obstacle-crossing process, θ0 and θ2 are the substation inspection robot obstacle-crossing parameters, F1 and F2 are the driving forces provided by the robot on the ground part and the obstacle part during the obstacle-crossing process, X is the distance from the contact point of the robot to the obstacle, α is the angle between the robot vehicle body axis and the ground, and i is the obstacle-crossing stage.

[0063] ​The multi-objective optimization model has two optimization objectives, the first optimization objective is to make the battery energy consumption in the process of obstacle crossing as small as possible, that is, W takes the minimum value as possible, and the second optimization objective is to make the driving torque peak value in the process of obstacle crossing as small as possible, that is, F takes the minimum value as possible. The optimization objectives are:

[0064]

[0065] The sequence and content of solving the multi-objective optimization model in the step S4 are as shown in Figure 5 , which is divided into 9 sub-processes:

[0066] S41: Set the specific obstacle height H, determine the value range of the individual according to the value range of the obstacle crossing parameter, and then initialize the population;

[0067] S42: Perform non-dominated sorting and crowding degree calculation on the initialized population, and each individual has its own rank and crowding degree.

[0068] S43: Set the iteration number, and stop iteration if the iteration number is exceeded;

[0069] S44: Compare the ranks of the individuals two by two, select the individual with a smaller rank, and if the ranks are the same, select the individual with a larger crowding degree,

[0070] S45: Obtain a new population by crossover, mutation and population merging;

[0071] S46: Calculate the current non-dominated sorting and crowding degree again;

[0072] S47: Update the population, and generate a new population through the elite strategy competition;

[0073] S48: Determine whether the maximum iteration number is reached, and if not, return to the fourth process to continue the steps S44-S48;

[0074] S49: The algorithm ends, and the Pareto front under each obstacle height H is obtained. Taking H=0.2m as an example, the Pareto front is as shown in Figure 6 .

[0075] Through the optimization work of the step S4, the Pareto front under each obstacle height H can be obtained, that is, a set of θ0 and θ2 values that achieve the smallest W and F under the current H, and the current values of W and F can also be obtained.

[0076] The sequence and content of establishing the obstacle crossing fuzzy neural network of the substation inspection robot in the step S5 are as shown in Figure 7 , which is divided into the following 3 sub-processes:

[0077] S51: Set network input nodes, hidden nodes and output nodes. The input nodes include three obstacle heights H, battery energy consumption W in the robot obstacle crossing process and driving torque peak value F in the obstacle crossing process. The output nodes include two first swing arm lifting angles θ0 and second swing arm lifting angles θ2;

[0078] S52: Normalize the Pareto frontiers in each obstacle height H in step S4, and use the normalized Pareto frontiers as samples of fuzzy rules in the fuzzy neural network. The samples are input into the fuzzy neural network for training.

[0079] S53: Train the fuzzy neural network to obtain the obstacle crossing parameters, i.e. the first swing arm lifting angle θ0 and the second swing arm lifting angle θ2, under any obstacle height H, battery energy consumption W and maximum driving force F, thereby obtaining the optimized obstacle crossing action sequence of the substation inspection robot under different obstacle crossing heights.

[0080] In summary, the above is only a preferred embodiment of the present application, and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A substation inspection robot obstacle crossing action sequence optimization method based on NSGA-II and fuzzy neural network, characterized in that, Comprise the following steps: S1: determine the substation inspection robot size parameters, step obstacle size parameters, including: car body length, wheel radius, car body mass, mass center position, front and rear swing arm length, swing arm wheel radius, swing arm mass, obstacle height; S2: establish a substation inspection robot obstacle dynamics model, the input variable is the substation inspection robot size parameters, obstacle height, obstacle parameter; the output result is the driving torque peak value in the obstacle process and the battery energy consumption in the obstacle process; S3: based on NAGA-Ⅱ algorithm, establish a substation inspection robot obstacle multi-objective optimization model; the independent variable of the optimization model is: 2 obstacle parameters θ0 and θ2; the optimization goal of the optimization model is: the driving torque peak value in the obstacle process is as small as possible, the battery energy consumption in the obstacle process is as small as possible; S4: use NSGA-Ⅱ algorithm to solve the multi-objective optimization model, thereby obtaining the multi-objective optimization Pareto frontier under different obstacle heights H; S5: establish a substation inspection robot obstacle dynamic fuzzy neural network, the fuzzy neural network input is the obstacle height, the battery energy consumption in the obstacle process, the driving torque peak value, the fuzzy neural network output is the obstacle parameter under the corresponding input condition, and the fuzzy neural network learning sample is the Pareto frontier; The substation inspection robot obstacle multi-objective optimization model in step S3 is specifically: Wherein, W is the battery energy consumption in the robot obstacle process, F is the driving torque peak value in the obstacle process, θ0 and θ2 are the substation inspection robot obstacle parameters, F1 and F2 are the driving force provided by the robot in the ground part and the obstacle part in the obstacle process, X is the distance from the robot contact point to the obstacle, α is the angle between the robot body axis and the ground, and i is the obstacle stage; The sequence and content of establishing a substation inspection robot obstacle fuzzy neural network in step S5 are divided into the following three sub-processes: S51: set the network input node, hidden node and output node; wherein the input node has three, which are obstacle height H, battery energy consumption W in the robot obstacle process and driving torque peak value F in the obstacle process; the output node has two, which are the first swing arm lifting angle θ0 and the second swing arm lifting angle θ2; S52: normalize each obstacle height H under the Pareto frontier in step S4, and use it as a sample of fuzzy rule in the fuzzy neural network, and input the sample into the fuzzy neural network for training; S53: train the fuzzy neural network to obtain the obstacle parameters under any obstacle height H, battery energy consumption W and maximum driving force F, that is, the first swing arm lifting angle θ0 and the second swing arm lifting angle θ2, so as to obtain the optimized substation inspection robot obstacle action sequence under different obstacle heights.

2. The substation inspection robot obstacle-crossing action sequence optimization method based on NSGA-II and fuzzy neural network according to claim 1, characterized in that: The obstacle dynamics model in step S2 is for the process from when the substation inspection robot contacts the obstacle with the front swing arm to when the mass center completely passes the obstacle angle line; wherein the obstacle parameter is the angle between the swing arm and the ground when the robot lifts the swing arm twice during the obstacle process.

3. The substation inspection robot obstacle-crossing action sequence optimization method based on NSGA-II and fuzzy neural network according to claim 1, characterized in that: The multi-objective optimization model has two optimization objectives. The first optimization objective is to minimize the battery energy consumption in the obstacle crossing process, i.e., W is as small as possible. The second optimization objective is to minimize the driving torque peak value in the obstacle crossing process, i.e., F is as small as possible. The optimization objectives are:

4. The substation inspection robot obstacle-crossing action sequence optimization method based on NSGA-II and fuzzy neural network according to claim 1, characterized in that: The order and content of solving the multi-objective optimization model in step S4 by using the NSGA-II algorithm are divided into 9 sub-processes: S41: Set the specific obstacle height H, determine the value range of the individual according to the value range of the obstacle crossing parameter, and then initialize the population; S42: Perform non-dominated sorting and crowding degree calculation on the initialized population, and each individual has its own rank and crowding degree; S43: Set the iteration number, and stop iteration if the iteration number is exceeded; S44: Compare the ranks of the individuals, select the individual with a smaller rank, and if the ranks are the same, select the individual with a larger crowding degree, S45: Obtain a new population by crossover, mutation and population merging; S46: Calculate the current non-dominated sorting and crowding degree again; S47: Update the population, and generate a new population by competition through the elite strategy; S48: Determine whether the maximum iteration number is reached, and if not, return to the fourth process to continue the steps S44-S48; S49: The algorithm is ended, and the Pareto front under each obstacle height H is obtained.

Images