Substation inspection robot motion control optimization method and system based on deep learning

Abstract

The invention discloses a power transformation inspection robot motion control optimization method and system based on deep learning, and the method comprises the steps: carrying out the diagnosis of the ill-conditioned condition of a normal matrix N0 based on the condition number of the normal matrix N0 in a loss function, carrying out the correction of the normal matrix through a Tikhonov regularization method, adjusting three minimum values in the feature values of the normal matrix, and carrying out the optimization of the motion control of a power transformation inspection robot. Obtaining a corrected normal matrix; constructing a path planning method and a reward function optimization method based on the corrected normal matrix, and designing neural network input and output based on deep learning; and motion optimization control of the power transformation inspection robot is realized. According to the method, the inspection robot can realize all-collision-free inspection point traversal path planning with the optimal path, and detailed reward and punishment evaluation is carried out on substation inspection tasks, collision trend, trend of approaching a new detection point, collision, reaching the new detection point, crossing the detection point, bypassing, timeout and the like.

Description

Background Art

[0002] Nowadays, electric energy has become an indispensable part of social life. From people's daily life to the progress of science and technology, the development of the country cannot be separated from electricity. Substation is one of the core facilities in the power system. Ensuring its safe and stable operation is a key issue in power safety production. Regular inspection of substation equipment is a basic work to ensure the normal operation of substation. Substation inspection is to inspect the main equipment in the substation to determine whether the equipment is in normal operation. After collecting relevant data, it is necessary to fill in the inspection report and record the inspection data. Therefore, the difference in the senses of operation and maintenance personnel makes the recorded data too subjective or incomplete. At the same time, substations usually occupy a large area and have a large number of internal equipment, resulting in a long inspection cycle, low work efficiency, and easy to miss inspections. The detected data cannot be entered into the management information system in a timely and accurate manner. Once the substation environment is too complex, it may bring safety hazards to the inspection work, especially under severe weather conditions. If the inspection cannot be carried out in time, the delay in time may turn a small problem into a major accident, leading to serious consequences. With the introduction of smart grids, the development trend of substations is large-scale, modernized, and complex. Therefore, traditional manual inspections can no longer meet the needs of substation inspections. A more intelligent inspection method can effectively reduce the workload of inspectors. With the rapid development of intelligent inspection robots, some repetitive and high-risk inspections can be gradually replaced by intelligent inspection robots to gradually replace manual inspections. The advantages of substation inspection robots are mainly reflected in repetitive work, which can be performed tirelessly. With advanced technology, inspection tasks can be flexibly formulated and regular inspections can be carried out. Therefore, it is an inevitable trend for the development of smart substations that intelligent inspection robots replace manual inspections.

[0003] Existing intelligent inspection robots use infrared, image, ultrasonic and other sensor technologies to detect faults in power plant equipment, automatically measure the actual temperature of the equipment, read the value of the instrument, identify the damage of the equipment, and send the collected data and test results to the control room in real time for further processing by power plant managers. It can automatically monitor the safety of power plant equipment all day, greatly reducing labor costs and error rates. In the inspection work of the intelligent substation inspection robot, it is necessary to check the docking position according to different map environments and power equipment, and plan a reasonable inspection path so that it passes through all parking points during the inspection process without colliding with obstacles in the environment, and at the same time make the robot's running path as short as possible. The current path planning method has made great progress in the field of robot navigation, but many of the latest path planning results have not been applied to the scene of power station inspection robots. Therefore, it is necessary to study the path planning of inspection robots in substation scenarios, which can provide feasible ideas for the design of domestic substation inspection robot systems.

[0004] Navigation and positioning is an important research direction for realizing autonomous work of inspection robots. Since the working environment of inspection robots is very complex, their working paths are often set near high-voltage transmission lines or buried underground cables. Strong electric fields and alternating magnetic fields are distributed around these transmission lines. Such a complex electromagnetic environment makes it particularly important to select high-precision and electromagnetic interference-resistant sensors. Nowadays, visual navigation technology is widely used in inspection robots due to its advantages such as high spatial resolution, accurate object recognition, and wide target recognition range. In addition, when inspecting high-voltage overhead lines, inspection robots specially developed for high-voltage overhead line inspections often sway in the wind under the action of wind. At this time, in order to balance its own center of gravity, the inspection robot will make subtle adjustments to its position, which may further aggravate the swing of the line. In this way, the difficulty of writing line obstacle removal methods is increased. The method of navigation using high-precision electromagnetic sensors can effectively reduce the interference of strong magnetic fields on detection signals in complex electromagnetic environments, and has the characteristics of reliable operation, easy installation, and good economy. However, in order to meet the requirements of navigation accuracy and detection sensitivity, a large number of sensors need to be installed to ensure data diversity, which also puts higher requirements on the performance of the sensors themselves. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a substation inspection robot motion control optimization method and system based on deep learning in response to the above-mentioned deficiencies in the prior art, which is used to solve the problem that the robot has a long detour path or fails to avoid obstacles in a complex substation environment; as well as technical problems that easily lead to speed fluctuations or trajectory deviations, and to improve positioning accuracy and speed.

[0006] The present invention adopts the following technical solutions:

[0007] The motion control optimization method of the substation inspection robot based on deep learning includes the following steps:

[0008] Based on the condition number of the normal matrix N0 in the loss function, the pathological condition of the normal matrix N0 is diagnosed, and the normal matrix is ​​corrected by the Tikhonov regularization method, and the three minimum values ​​of the eigenvalues ​​of the normal matrix are adjusted to obtain the corrected normal matrix.

[0009] Based on the corrected normal matrix, the path planning method and reward function optimization method are constructed, and the neural network input and output based on deep learning are designed to realize the motion optimization control of the substation inspection robot.

[0010] Preferably, the condition number based on the normal matrix N0 is as follows:

[0011]

[0012] Among them, λ1 is the maximum eigenvalue of matrix N0, λ m is the smallest non-zero eigenvalue of the matrix N0.

[0013] 3. The motion control optimization method for substation inspection robot based on deep learning according to claim 2 is characterized in that the spectral decomposition of the normal matrix N0 is:

[0014]

[0015] Among them, Q is an orthogonal matrix.

[0016] Preferably, the ill-conditioned model is corrected by using the Tikhonov regularization method, and the ill-conditioned model after linearization is set with a criterion function, and the ill-conditioned model is solved to obtain a stable solution Z of the ill-posed problem.

[0017] Preferably, the single-difference carrier phase observation equation is solved using WLS and Tikhonov regularization, and after obtaining the floating-point solution and covariance matrix of the single-difference periodic ambiguity, the LAMBDA algorithm is used to implement fixed ambiguity to obtain the accurate position information of the receiver.

[0018] Preferably, in step S2, the input-output mapping of the neural network is specifically:

[0019] The historical actions of the inspection robot and the interaction information between the inspection robot's own sensors and the environment are taken as the input of the neural network, and the angular velocity of the left and right wheels of the inspection robot is taken as the output of the neural network. Through the positive and negative rewards of the collision behavior of the reward function, the inspection robot is induced to reach the goal of crossing the inspection point and avoiding collision.

[0020] Preferably, the reward function is as follows:

[0021] When r-0 (t) <d nearcol :

[0022]

[0023] When the inspection robot arrives at a new inspection point:

[0024] r(a t ,s t )=r ste n ste +L str r mov +r new n new

[0025] Time it takes for the inspection robot to complete the checkpoint traversal:

[0026] r(a t ,st )=r ste n ste +L str r mov +2r all

[0027] When the inspection robot collides with an obstacle:

[0028] r(a t ,s t )=r ste n ste +L str r mov +r col

[0029] Among them, r ste is the negative ste reward for the number of steps the robot takes, n ste is the number of steps the robot moves, r mov is the negative reward mov for the robot’s moving distance, L str is the linear distance the robot moves from point A to point B; r nearco is the negative reward when the robot approaches collision nearcol, is the distance between the robot and the nearest obstacle r-0 at time t; d is the distance when the robot approaches the collision point, n new is the number of new detection points reached by the new robot; r new The reward for the robot to reach a new detection point; r mov is the reward for reaching all detection points; r col Negative reward col for collision.

[0030] Preferably, the first control variable of the neural network architecture is the default value of the data packet, the structure of the three hidden layers each contains 32 neurons, and the multilayer perceptron neural network is connected by a function F(x), and the function F(x) is composed of multiple linear functions and nonlinear activation functions. The MLP is expressed as:

[0031] F(x)=λ n ·σ n-1 ·λ n-1 ...λ1(x)·S

[0032] Among them, λ n represents each layer i in the neural network, σ is the activation function, S is the softmax function, and the MLP value n of each neuron i,j for:

[0033] n i,j =f(∑ j w j ·n i-1,j +bj )

[0034] Among them, w j is the connection weight of each layer in the neural network, b j is the bias value of the neuron, f is the activation function, n i-1,j is the output value of the jth neuron in the i-1th layer of the neural network.

[0035] Preferably, the reward function r(a t ,s t )as follows:

[0036]

[0037] Among them, d r-t (t) is the distance from the rt robot to the target detection point at time t; r new is the reward for the new robot to reach the new detection point; r ste is the negative reward ste for the number of steps the robot takes, n ste is the number of steps the robot moves ste, r mov is the negative reward for the robot moving the mov distance, L str is the linear distance the robot moves from point A to point B; r nearcol is the negative reward when the robot nearcol approaches a collision, d r-o (t) is the distance between the robot and the nearest obstacle r-0 at time t; r all The reward for all to reach all checkpoints; r col is the negative reward when col collision occurs, r out is the negative reward when timeout occurs, d nearcol is the distance between the robot nearcol and the obstacle when it is close to collision, r app is the distance when the app robot approaches the detection point, d win is the distance that the robot reaches the detection point at which it is considered to have won; n new is the number of new detection points reached by the robot.

[0038] In a second aspect, an embodiment of the present invention provides a substation inspection robot motion control optimization system based on deep learning, comprising:

[0039] The correction module diagnoses the pathological condition of the normal matrix N0 based on the condition number of the normal matrix N0 in the loss function, corrects the normal matrix through the Tikhonov regularization method, adjusts the three minimum values ​​of the eigenvalues ​​of the normal matrix, and obtains the corrected normal matrix;

[0040] The control module constructs the path planning method and reward function optimization method based on the corrected normal matrix, designs the neural network input and output based on deep learning, and realizes the motion optimization control of the substation inspection robot.

[0041] In a third aspect, a computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned deep learning-based substation inspection robot motion control optimization method when executing the computer program.

[0042] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned substation inspection robot motion control optimization method based on deep learning.

[0043] Compared with the prior art, the present invention has at least the following beneficial effects:

[0044] A motion control optimization method for substation inspection robots based on deep learning is proposed. A high-precision positioning method for transformer inspection robots is proposed. This method adopts a single base station, dual-frequency, dual-antenna, and four-signal structure, and uses pseudo-range observation and carrier phase observation for high-precision positioning solutions. The Tikhonov regularization is used to correct the model's pathological nature. Substation inspection requires the traversal of equipment inspection points based on public path planning. A reward function with multiple scoring rules is designed, which can enable the inspection robot to achieve collision-free inspection point traversal path planning with the optimal path. For substation inspection tasks, detailed reward and penalty evaluations are performed on collision trends, trends approaching new detection points, collisions, arrival at new detection points, crossing detection points, detours, and timeouts. The superiority of the method of the present invention is proved through comparison and simulation experiments, and high-precision positioning can be achieved.

[0045] Furthermore, the purpose of setting the condition number based on the normal matrix is ​​to evaluate and improve the numerical stability of the matrix, thereby improving the accuracy and robustness of the calculation process. By calculating the condition number, you can intuitively judge whether the matrix is ​​close to singularity or pathological. The larger the condition number, the more likely the matrix will cause instability and inaccuracy in numerical calculations. Therefore, the condition number can be used as a tool to evaluate the numerical stability of the matrix. By adjusting the eigenvalues ​​of the matrix, the condition number can be reduced, thereby improving the stability of the calculation. The eigenvalue distribution of the normal matrix directly affects its condition number. Assume that the eigenvalues ​​of the matrix N0 vary greatly (that is, λ1 is large, and λ m This indicates that N0 may extremely stretch or compress the input vector in some directions, resulting in increased sensitivity of the numerical calculation.

[0046] Furthermore, when the condition number is large, the Tikhonov regularization method can be used to correct the matrix. Tikhonov regularization adjusts the eigenvalues ​​by adding a small positive value to the diagonal of the matrix, thereby reducing the condition number and improving the numerical stability of the matrix. By properly selecting α, the condition number can be effectively reduced, making the calculation more stable.

[0047] Furthermore, the weighted least squares (WLS) and Tikhonov regularization method are used to solve the single-difference carrier phase observation equation, and the LAMBDA algorithm is combined to fix the ambiguity to obtain the precise position, which is a high-precision positioning technology; the carrier phase observation equation is more accurate than the pseudo-range observation equation, but it also introduces the ambiguity problem; solving it with WLS and Tikhonov regularization and using the LAMBDA algorithm to fix the ambiguity can significantly improve the positioning accuracy. Tikhonov regularization stabilizes the solution process by introducing additional constraints, avoids numerical instability caused by the ill-conditioned observation equation or excessive condition number, and enhances the stability and robustness of the solution. In the single-difference observation equation, each observation may have different noise levels, so the WLS method is used to solve the floating-point solution of the ambiguity by minimizing the weighted residual sum of squares; when the condition number of the observation matrix AAA is large, direct solution may lead to unstable results. Tikhonov regularization improves the stability of the solution by introducing a regularization term; Tikhonov regularization effectively reduces the impact of ill-conditioned problems and improves the stability of the ambiguity solution.

[0048] Furthermore, the input-output mapping setting of the neural network is crucial in the machine learning model. Its main purpose is to optimize the model training process through the design of reasonable input features and target outputs, thereby improving the model's prediction and generalization capabilities. This is achieved by optimizing the expression of feature information and the accuracy of the task. The ultimate goal is to enable the neural network to learn the complex relationship between input data and output targets more efficiently.

[0049] Furthermore, by setting multiple hidden layers and 32 neurons in each layer, the network can extract and process information at different levels, thereby enhancing the functionality and adaptability of the model; the activation function introduces nonlinear characteristics, allowing the network to handle more complex features and patterns. Without nonlinear activation functions, a network with multiple layers is equivalent to a linear model and cannot handle complex nonlinear problems. Nonlinear activation functions enable the network to fit complex functional relationships and enhance the network's expressiveness and learning capabilities. By combining linear transformations and nonlinear activation functions, MLP can perform nonlinear transformations on data, thereby learning more complex patterns and features in the data.

[0050] It can be understood that the beneficial effects of the second aspect mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0051] To sum up, the present invention enables the inspection robot to realize collision-free inspection point traversal path planning with the optimal path. For substation inspection tasks, detailed reward and penalty evaluation is carried out for collision trends, trends approaching new inspection points, collisions, arrival at new inspection points, crossing inspection points, detours, timeouts, etc.

[0052] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of the high-precision positioning method of the substation inspection robot of the present invention;

[0054] Figure 2 This is a historical motion analysis diagram of the robot of the present invention;

[0055] Figure 3 The updated neural network input-output mapping diagram of the present invention;

[0056] Figure 4 This is a simulation result diagram of the GDOP distribution in the positioning area of ​​the present invention;

[0057] Figure 5 This is a simulation result diagram of HDOP distribution in the positioning area of ​​the present invention;

[0058] Figure 6 This is a simulation result diagram of VDOP distribution in the positioning area of ​​the present invention;

[0059] Figure 7 It is the error diagram of simulation results of all observation epochs of the present invention;

[0060] Figure 8 is a comparison chart of the overall error results of the five methods of the present invention;

[0061] Fig. 9 It is the robot motion speed accuracy diagram of different methods of the present invention;

[0062] Fig.10 is the stability diagram of the robot climbing path under the five methods of the present invention;

[0063] Fig.11 Path planning and navigation roadmap for the present invention;

[0064] Fig.12 A schematic diagram of a computer device provided by an embodiment of the present invention;

[0065] Fig.13The present invention is a block diagram of an electronic device provided according to an embodiment. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] In the description of the present invention, it should be understood that the terms “include” and “comprises” indicate the presence of described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0068] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0069] It should be further understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present invention generally indicates that the associated objects are in an "or" relationship.

[0070] It should be understood that, although the terms first, second, third, etc. may be used to describe preset ranges, etc. in the embodiments of the present invention, these preset ranges should not be limited to these terms. These terms are only used to distinguish preset ranges from each other. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0071] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0072] Various structural schematic diagrams of the embodiments disclosed in the present invention are shown in the accompanying drawings. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clear expression. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0073] The present invention provides a motion control optimization method for a substation inspection robot based on deep learning, which adopts a single base station, dual-frequency, dual-antenna, and four-signal structure, and combines pseudo-range observation and carrier phase observation for positioning; the positioning accuracy is improved, and compared with the traditional single-frequency or single-antenna technology, the accuracy and stability of positioning can be significantly improved; the Tikhonov regularization is introduced to correct the pathological nature of the model; this is an effective method for solving the problems of model instability and overfitting in high-dimensional data; by reducing the pathological nature of the model, the stability of the positioning algorithm is enhanced, the positioning error is reduced, and the reliability of the model in practical applications is improved; a reward function with multiple scoring rules that comprehensively considers factors such as collision trend, approaching a new detection point, collision, reaching a new detection point, crossing a detection point, detouring, and timeout is designed; the reward function can comprehensively evaluate and optimize path planning, which not only improves the efficiency of the inspection robot, but also improves safety, reduces collisions and detours in path planning, and ensures the smooth completion of the inspection task; through the above improvements, the present invention is significantly superior to the prior art in terms of positioning accuracy, model stability, path planning optimization, and system safety.

[0074] See also Figure 1 The present invention provides a motion control optimization method for a substation inspection robot based on deep learning, comprising the following steps:

[0075] S1. Study the rapid positioning method model of the ground-based pseudo-satellite navigation and positioning system, analyze the pathological problems of the positioning algorithm model in the rapid positioning method, and propose a high-precision positioning method for the transformer inspection robot;

[0076] The rapid positioning method model of the ground-based pseudo-satellite navigation and positioning system is studied, the pathological problems of the rapid positioning method model are analyzed, and the steps of the high-precision positioning method of the transformer inspection robot are proposed, including:

[0077] S101. Analyze the pathological problems of the sick model

[0078] An ill-conditioned model is one where a very small perturbation of a parameter leads to a drastic change in the result.

[0079] In the positioning solution process of the ground-based pseudo-satellite navigation and positioning system, the observation parameters are generally solved in a shorter time to improve the solution speed. However, due to the small sampling interval, the spatial geometry of the base station and the receiver does not change much, which is similar to the insufficient observation information caused by repeated observations. This leads to a strong correlation between the data columns of the coefficient matrix of the observation model, that is, there is complex collinearity in the coefficient matrix, which will affect the normal matrix and lead to an ill-conditioned model. According to the steps of the navigation and positioning method, after obtaining the floating-point solution of the integer ambiguity, the LAMBDA method is used to fix the floating-point solution. However, the LAMBDA method can only accurately and effectively determine the precise floating-point solution with the corresponding covariance matrix. When the solution model is ill-conditioned, LAMBDA (Least-squares AM Biguity Decorrelation Adjustment) will bring a large solution error, thereby affecting the solution of the LAMBDA method, resulting in unsatisfactory positioning solution results, resulting in the ill-conditioned model of the ground-based pseudo-satellite navigation and positioning system's fast positioning solution will have a great impact on the positioning accuracy;

[0080] S102, diagnose the pathological model;

[0081] For the linear system N 0x =z, assuming that the observed quantity z is accurate, and the coefficient matrix N0 has an error ΔN0, similarly, the solution of the equation system is X+Δx;

[0082] (ΔN0+N0)(Δx+x)=Z(1)

[0083] Known N 0x =z, then:

[0084]

[0085] Then we get:

[0086]

[0087] Even if the solution process is completely correct, whether the change in the result caused by some data errors is controllable depends on the size of the quantity ||N||*||N -1 ||, when the condition number of the normal matrix N0 is large (the normal matrix is ​​ill-conditioned), even if N0 and z have small errors, it may cause huge solution deviations, resulting in large errors in the subsequent use of the LAMBDA method to fix the floating-point solution, or even cause the floating-point solution to fail to fix;

[0088] Spectral decomposition of the normal matrix N0:

[0089]

[0090] Among them, Q is an orthogonal matrix; get N -1The spectral decomposition of , we get the condition number of the normal matrix N0:

[0091]

[0092] S103. Training the pathological model

[0093] The condition number indicates the degree of numerical variation of the N0 eigenvalues. The condition number is used to determine the pathology of the pathological model, that is, the diagnosis of the pathology is determined by the condition number of the normal matrix;

[0094] The ill-conditioned model is corrected by using the Tikhonov regularization method. The linearized ill-conditioned model is:

[0095] C·Z=L(6)

[0096] Among them, L is the known observation, C is the coefficient matrix, and Z is the parameter to be estimated. According to the calculation principle of WLS, the solution of the above formula is:

[0097] Z=C T LP n (C T CP n ) -1 (7)

[0098] In order to obtain a stable solution to the ill-posed problem, a criterion function is set:

[0099]

[0100] The parameter Z that minimizes the above equation is a stable solution. In the above equation, P is the regularization matrix, which is a regularization parameter and plays a balancing role. By minimizing the above formula, we get:

[0101] Z=P n C T L(αP Z +C T CP n ) -1 (9)

[0102] The determination of the regularization matrix PZ is the key to Tikhonov regularization. After selecting the regularization matrix PZ, the value of the smoothing factor is tested, and finally **=1 is selected. Through Tikhonov regularization, the three minimum values ​​of the normal matrix eigenvalues ​​are adjusted so that they are no longer much smaller than other eigenvalues, and the pathological condition of the regularized model is greatly improved.

[0103] S104. A high-precision positioning method for a transformer inspection robot used in a ground pseudo-satellite navigation and positioning system.

[0104] See also Figure 1Each base station has two independent transmitting antennas, and each antenna transmits two navigation signals with frequencies of f1 (2.412GHz) and f2 (2.461GHz), that is, a single base station can transmit a total of 4 navigation signals. The advantages are: two transmitting antennas are equivalent to two base stations, which can improve the geometry of the receiver and the base station, and have a better GDOP distribution than a single antenna, thereby improving the positioning accuracy of the system; within the same observation time, more carrier phase observations are added, and under the same observation conditions, the receiver's solution speed and positioning accuracy can be improved.

[0105] In order to achieve rapid positioning, the receiver needs to solve the position within a shorter observation epoch, which will lead to serious complex collinearity in the observation equation, resulting in the ill-conditioning of the solution model and affecting the accuracy of the floating-point solution. Therefore, in the process of solving the single-difference carrier phase observation equation, the Tikhonov regularization is used to correct the ill-conditioning of the normal matrix, which can solve the position of the receiver within a shorter observation epoch, thereby providing rapid positioning while improving the positioning accuracy.

[0106] The single-difference carrier phase observation equation is solved by WLS and Tikhonov regularization. After obtaining the floating-point solution and covariance matrix of the single-difference periodic ambiguity, the LAMBDA method is used to achieve fast and accurate fixed ambiguity, thereby obtaining the accurate position information of the receiver.

[0107] In order to achieve rapid positioning, the receiver needs to solve the position within a shorter observation epoch, which will lead to serious complex collinearity in the observation equation, resulting in the ill-conditioning of the solution model and affecting the accuracy of the floating-point solution. Therefore, in the process of solving the single-difference carrier phase observation equation, the Tikhonov regularization is used to correct the ill-conditioning of the normal matrix, which can solve the position of the receiver within a shorter observation epoch, thereby providing rapid positioning while improving the positioning accuracy.

[0108] The single-difference carrier phase observation equation is solved by WLS and Tikhonov regularization. After obtaining the floating-point solution and covariance matrix of the single-difference periodic ambiguity, the LAMBDA method is used to achieve fast and accurate fixed ambiguity, thereby obtaining the accurate position information of the receiver.

[0109] The calculation process is as follows: In a certain observation epoch, the receiver obtains the coarse position information of the receiver through pseudo-range observation. The pseudo-range observation equation obtained by the receiver by receiving the navigation signal of base station i is as follows:

[0110] p i =cr i +δt u +ε i +T i (10)

[0111] Where p is the pseudo distance measured by the receiver; r is the actual physical distance between the receiver and the base station transmitting antenna; δt u represents the clock error of the receiver; T represents the tropospheric propagation error during signal propagation, which is corrected by a high-precision tropospheric error model; ε refers to other unquantified measurement errors, including multipath error and receiver thermal noise error.

[0112] Differential technology is widely used in precision navigation and positioning systems to eliminate some common errors and improve the positioning accuracy of the system. Assume that the receiver receives navigation signals from m+1 base stations at the same time, that is, there are 2m+2 pseudo-range observation values. The signal transmitted by the two antennas of base station No. 1 is used as the reference signal, and the pseudo-range single difference observation equation is obtained:

[0113]

[0114] The tropospheric propagation error corrected by the high-precision tropospheric error model is used as the pseudo-range measurement error. The unknown parameters in the above equation are only the position coordinates of the receiver. When the number of observation equations 2m is greater than or equal to 3, the position coordinates (x, y0, z0) of the user receiver are obtained and substituted into the carrier phase observation equation as the initial estimate to solve it. The high accuracy of the initial estimate is conducive to the rapid convergence of the method. The carrier phase observation equation of the ground-based pseudo-satellite navigation and positioning system is as follows:

[0115] φ=λ(c·T+c·δt u +N+r)+ε φ (12)

[0116] The receiver receives the navigation signals from base stations i and j at one end, and takes base station j as the reference base station, and obtains the single-difference carrier phase observation equation:

[0117] Δφ ij =λ -1 (r ij +T ij +v ij )+N ij (13)

[0118] Among them, r ij is the distance difference between the receiver and base station I, j; T ij represents the difference in tropospheric propagation delay; N ij represents the single-difference cyclic ambiguity; v ij In order to measure the difference of noise error, the receiver position coordinates obtained by pseudo-range (x, y, z) are used as the initial estimate of the WLS method, and the 4mn dual-frequency single-difference carrier phase observation equation is linearized to obtain the linear observation equation as follows:

[0119] Δz=ANN Δx N +A CC Δx c +ξ(14)

[0120] Where Δz is the difference between the known observation and the estimated parameter; A CC is the Jacobian matrix of the single-differenced circular fuzzy; Δx c is the difference between the receiver position coordinates and the initial estimate; A NN is the Jacobian matrix of the single-differenced circular fuzzy, Δx N is the difference between the dual-frequency single-difference cyclic ambiguity and the initial estimate; ξ is the observation error.

[0121] The difference between the receiver position coordinates and the initial estimate and the floating point solution of the dual-frequency single-difference cyclic ambiguity are:

[0122]

[0123] After multiple iterations, the floating-point solutions of the position coordinates and the dual-frequency single-difference cyclic ambiguity and the corresponding error covariance matrix are obtained until the iterative results converge.

[0124] The core of the LAMBDA method consists of two parts: the correlation processing of fuzzy reduction and the integer solution search of integer ambiguities. The LAMBDA method first performs integer transformation on the integer ambiguities to obtain a new weakly correlated floating-point ambiguity solution, and then performs a sequential conditional least squares search to obtain a fixed solution for the integer ambiguities.

[0125] The main process of fuzzy reduction related processing is to convert the multidimensional ellipsoid into a spherical ellipsoid by performing integer Gaussian transformation on the circular ambiguity covariance matrix line, which can greatly improve the search efficiency. The floating point solution has different differences from the actual data, while the integer solutions fixed by the LAMBDA method are equal to the actual data.

[0126] S2. Construct path planning methods and reward function optimization methods, and the design process of neural networks, including the design of neural network input and output, the selection of neural network types, comparative analysis of the number of hidden layers, comparative analysis of the number of neurons in each hidden layer, and comparative analysis of activation functions;

[0127] S201. Design the input-output mapping of the neural network;

[0128] In the experiment, the movement of the inspection robot is achieved by controlling the output power of the drive motor. However, there is a certain error in converting the output power of the inspection robot drive motor into the output angular velocity of the tire. Therefore, in the simulation, the output angular velocity of the left and right wheels of the inspection robot is directly used as the control object. In the design of the neural network input-output mapping, the angular velocity of the left and right wheels of the inspection robot is used as the output of the neural network.

[0129] Although the position of the detection point is completely unknown before the detection robot starts to cross the detection point, it can use the interaction information between its own sensors and the environment as the input of the neural network, and induce the detection robot to reach the goal of crossing the detection point and avoiding collision through the positive and negative rewards of the collision behavior of the reward function.

[0130] The detection point is set as the signal transmitter, and the detection robot receives the signal within the signal radiation range. In the simulation experiment, the obstacle attenuation and reflection obstruction of the signal are ignored.

[0131] The black rectangles are obstacles, the measuring points are represented by stars, the spatial distance range detected by the solid-state laser radar is set to (0,20), and the robot is equipped with 4 solid-state laser radars in a circular arrangement. The signals emitted by the radars will be reflected and received when encountering obstacles, so that the distances to the obstacles in four directions can be detected.

[0132] Mainly targeting two-wheeled inspection robots, the robot's state transfer is achieved through the change of the angular velocity of the left and right wheels. After determining the state input and behavior output of the neural network, a reward function is designed according to the principles of reward function design and the characteristics of the substation inspection task. The reward function must have the function of giving positive rewards to the inspection robot's behavior of crossing the inspection point and negative rewards to the inspection robot's collision behavior. Several commonly used neural network models, hidden layer numbers and activation functions are selected to study whether the design of the neural network input-output mapping can help the inspection robot achieve the goal of completely collision-free.

[0133] The reward function used for simulation and comparative experiments is as follows:

[0134] When r-0 (t) <d nearcol :

[0135]

[0136] When the inspection robot arrives at a new inspection point:

[0137] r(a t ,s t )=r ste n ste +L str r mov +r new n new (17)

[0138] Time it takes for the inspection robot to complete the checkpoint traversal:

[0139] r(a t ,s t )=rste n ste +L str r mov +2r all (18)

[0140] When the inspection robot collides with an obstacle:

[0141] r(a t ,s t )=r ste n ste +L str r mov +r col (19)

[0142] Among them, r ste is the negative ste reward for the number of steps the robot takes, n ste is the number of steps the robot moves, r mov is the negative reward mov for the robot’s moving distance, L str is the linear distance the robot moves from point A to point B; r nearco is the negative reward when the robot approaches collision nearcol, is the distance between the robot and the nearest obstacle r-0 at time t; d is the distance when the robot approaches the collision point, n new is the number of new detection points reached by the new robot; r new The reward for the robot to reach a new detection point; r mov is the reward for reaching all detection points; r col Negative reward col for collision.

[0143] The first control variable of the neural network architecture is the default value of the data packet. The structure of the three hidden layers contains 32 neurons each. The multilayer perceptron neural network is connected by a function F(x), which consists of multiple linear functions and nonlinear activation functions. The function has the characteristics of high computational efficiency, bounded derivatives, and easy derivation. The MLP is expressed as:

[0144] F(x)=λ n ·σ n-1 ·λ n-1 ...λ1(x)·S(20)

[0145] Among them, λ n represents each layer i in the neural network, σ is the activation function, S is the softmax function, and the MLP value of each neuron is:

[0146] n i,j =f(∑ j w j ·n i-1,j +b j)(twenty one)

[0147] Among them, w is the connection weight of each layer in the neural network, b is the bias value of the neuron, f is the activation function, the specific values ​​of w and b are encapsulated in the package, and the neurons in the current layer are functions of the output value of each connected neuron in the previous layer.

[0148] See also Figure 2 ,The robot’s historical actions are added as the fourth input of the neural network, in order to provide the robot with more sufficient information and help the inspection robot traverse the inspection points.

[0149] The robot has recorded several historical actions. Mobile robots and humans have different ways of thinking. For mobile robots, as long as there is a logical relationship between the input and output of the neural network design, the internal structure of the neural network is determined by training a large number of samples. When new information is input, the corresponding output can be generated according to the logic. Now add the historical actions of the inspection robot in the first two steps as the fourth input of the neural network. The updated input-output mapping diagram is as follows: Figure 3 shown.

[0150] S202, Reinforcement learning reward function analysis;

[0151] In reinforcement learning, the inspection robot's goal is formalized as a special signal, called gain, which is transmitted to the inspection robot through the environment. On each end, the gain is a single scalar value. The inspection robot's goal is to maximize the total gain it receives. This means that what needs to be maximized is not the current gain, but the long-term cumulative gain, maximizing the probability expectation of the cumulative sum of the scalar signals received by the agent (called the gain). Using gain signals to formalize goals is one of the most prominent features of reinforcement learning.

[0152] The inspection robot needs to learn how to maximize the reward. If the inspection robot is to be given a clear goal, the benefits must be provided in such a way that the agent can achieve the predetermined goal while maximizing the benefits. Therefore, the benefits should be set in a way that truly indicates what the security robot expects to achieve.

[0153] In particular, the reward signal does not impart prior knowledge about how the inspection robot achieves its goals.

[0154] The reward signal should only be used to communicate the goal to be achieved, not how to achieve it. Reward design is actually a key element in the Markov decision process, which is determined by the state, action, and next state. In actual engineering, reward design is very difficult and often does not achieve the expected results. In most cases, parameters need to be carefully adjusted to prevent the mobile robot from accidentally discovering the scoring technique. In actual engineering, even if there is no such bad situation as the agent's score, in order to achieve the effect of the method, it is necessary to constantly adjust the reward of the environment.

[0155] When the characteristics of the reinforcement learning system are not clear enough, the design of the reward function to induce the target behavior becomes more difficult. In addition, during the process of reward shaping, the formation of the reward is very important to the effect of reinforcement learning speed.

[0156] It can be seen that in the reward design work, reasonably setting the reward function to help the inspection robot achieve its goals is the top priority for the robot to complete the task of crossing the inspection point.

[0157] In order to allow the agent to act randomly at the beginning, the probability of bringing negative rewards and the probability of bringing positive rewards are approximately equal to 50%. In this way, the agent can perform reinforcement learning more efficiently. Compared with sparse rewards and formal rewards, this is also a scientific reward function design method.

[0158] S203. Study the reward function optimization method for substation inspection tasks.

[0159] The reward function can score each walk of the robot and adjust the internal weights of the neural network based on the high scores through proximal strategy optimization. Based on the existing reward function design method, the problems of sparse reward design and formal reward design are avoided. By drawing on the advantages of the variable reward coefficient reward design method, the goals of completing detection point traversal and avoiding obstacle collision are discussed in different cases, and negative rewards are handled according to other problems that may arise in the simulation scenario.

[0160] The designed reward function is as follows:

[0161]

[0162] Among them, d r-t (t) is the distance from the rt robot to the target detection point at time t; r new is the reward for the new robot to reach the new detection point; r ste is the negative reward ste for the number of steps the robot takes, n ste is the number of steps the robot moves ste, r mov is the negative reward for the robot moving the mov distance, L str is the linear distance the robot moves from point A to point B; rnearcol is the negative reward when the robot nearcol approaches a collision, d r-o (t) is the distance between the robot and the nearest obstacle r-0 at time t; r all The reward for all to reach all checkpoints; r col is the negative reward when col collision occurs, r out is the negative reward when timeout occurs, d nearcol is the distance between the robot nearcol and the obstacle when it is close to collision, r app is the distance when the app robot approaches the detection point, d win is the distance that the robot reaches the detection point at which it is considered to have won; n new is the number of new detection points reached by the robot.

[0163] Although the control goal is to make the inspection robot complete traversal and obstacle avoidance, the route to traverse the inspection points should also be more efficient. If the inspection robot avoids obstacles and walks in reverse several times, it will eventually complete the task of crossing the inspection points, but it is also meaningless;

[0164] For passing through new checkpoints and critically approaching new checkpoints, a large amount of rewards is set to encourage the robot to continue pursuing to reach the checkpoints. The positive rewards for reaching new checkpoints combined with the negative rewards brought by robot movement can both encourage the robot to continue to pursue reaching new checkpoints and encourage the robot to take the shortest path to get rewards faster;

[0165] Negative rewards for collisions and near collisions induce the robot's obstacle avoidance effect. In the case of timeout, negative rewards are set to make the robot work more efficiently. Combined with negative rewards for movement, the robot learns to plan more efficient movement routes to traverse the detection points. The maximum reward setting for traversal clearly defines the robot's final goal. If only this reward is set, it will fall into the misunderstanding of sparse rewards. The proposed reward function design method sets rewards for the robot to approach and reach new detection points, making the reward setting more dense and helping the robot to achieve the final goal in a gradient from easy to difficult. The design is to allow the robot to continuously traverse new detection points and avoid collisions, and ultimately achieve the goal of planning an optimal path that avoids collisions and traverses all detection points.

[0166] It will be appreciated by those skilled in the art that various aspects of the present invention may be implemented as systems, methods or program products. Therefore, various aspects of the present invention may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits", "modules" or "platforms".

[0167] In yet another embodiment of the present invention, a substation inspection robot motion control optimization system based on deep learning is provided. The system can be used to implement the above-mentioned substation inspection robot motion control optimization method based on deep learning. Specifically, the substation inspection robot motion control optimization system based on deep learning includes a correction module and a control module.

[0168] Among them, the correction module diagnoses the pathological state of the normal matrix N0 based on the condition number of the normal matrix N0 in the loss function, corrects the normal matrix through the Tikhonov regularization method, adjusts the three minimum values ​​of the eigenvalues ​​of the normal matrix, and obtains the corrected normal matrix;

[0169] The control module constructs the path planning method and reward function optimization method based on the corrected normal matrix, designs the neural network input and output based on deep learning, and realizes the motion optimization control of the substation inspection robot.

[0170] In another embodiment of the present invention, a terminal device is provided, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the substation inspection robot motion control optimization method based on deep learning, including:

[0171] The pathological condition of the normal matrix N0 is diagnosed based on the condition number of the normal matrix N0 in the loss function, and the normal matrix is ​​corrected by the Tikhonov regularization method. The three minimum values ​​of the eigenvalues ​​of the normal matrix are adjusted to obtain the corrected normal matrix. Based on the corrected normal matrix, the path planning method and reward function optimization method are constructed, and the neural network input and output based on deep learning are designed; the motion optimization control of the substation inspection robot is realized.

[0172] In another embodiment of the present invention, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device, and can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that more specific examples (non-exhaustive list) of the computer-readable storage medium here include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0173] Computer readable storage media also include data signals propagated in baseband or as part of a carrier wave, which carry readable program codes. Such propagated data signals can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable storage medium can also be any readable medium other than a readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or device. The program code contained on the readable storage medium can be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0174] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0175] The processor may load and execute one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the substation inspection robot motion control optimization method based on deep learning in the above embodiment; one or more instructions in the computer-readable storage medium are loaded and executed by the processor as follows:

[0176] The pathological condition of the normal matrix N0 is diagnosed based on the condition number of the normal matrix N0 in the loss function, and the normal matrix is ​​corrected by the Tikhonov regularization method. The three minimum values ​​of the eigenvalues ​​of the normal matrix are adjusted to obtain the corrected normal matrix. Based on the corrected normal matrix, the path planning method and reward function optimization method are constructed, and the neural network input and output based on deep learning are designed; the motion optimization control of the substation inspection robot is realized.

[0177] See also Fig.12 , the terminal device is a computer device. The computer device 60 of this embodiment includes: a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When the computer program 63 is executed by the processor 61, the substation inspection robot motion control optimization method based on deep learning in the embodiment is implemented. To avoid repetition, it is not described here one by one. Alternatively, when the computer program 63 is executed by the processor 61, the functions of each model / unit in the substation inspection robot motion control optimization system based on deep learning in the embodiment are implemented. To avoid repetition, it is not described here one by one.

[0178] The computer device 60 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will appreciate that Fig.12 This is only an example of the computer device 60 and does not constitute a limitation of the computer device 60. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, etc.

[0179] The processor 61 may be a central processing unit (CPU), other general-purpose processors, central processing units, graphics processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, data processing logic based on quantum computing, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0180] The memory 62 may be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 may also be an external storage device of the computer device 60, such as a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 60.

[0181] Furthermore, the memory 62 may include both an internal storage unit of the computer device 60 and an external storage device. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 may also be used to temporarily store data that has been output or is to be output.

[0182] Any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0183] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0184] See also Fig.13 The terminal device 600 is an electronic device, and the electronic device is in the form of a general computing device. The components of the electronic device may include but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including the storage unit 620 and the processing unit 610), a display unit 640, etc.

[0185] The storage unit stores program codes, which can be executed by the processing unit 610, so that the processing unit 610 performs the steps according to various exemplary embodiments of the present invention described in the above method section of this specification. For example, the processing unit 610 can perform the following steps: Figure 1 Follow the steps shown in .

[0186] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 6201 and / or a cache memory unit 6202 , and may further include a read-only memory unit (ROM) 6203 .

[0187] The storage unit 620 may also include a program / utility 6204 having a set (at least one) of program modules 6205, such program modules 6205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0188] Bus 630 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0189] The electronic device 600 may also communicate with one or more external devices 700 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 600, and / or communicate with any device that enables the electronic device 600 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 650. Furthermore, the electronic device 600 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 660. The network adapter 660 may communicate with other modules of the electronic device 600 via a bus 630. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms, etc.

[0190] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention described and shown in the drawings here can usually be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0191] Example 1

[0192] The method of the present invention is verified by simulation experiments, and two simulation schemes of single base station dual signals and single base station four signals are compared, and two schemes of using or not using Tikhonov regularization are simulated respectively;

[0193] The simulation verification experiment is divided into two schemes for comparison. Each scheme simulates the situation before and after using Tikhonov regularization to verify the advantages of the high-precision positioning method of the transformer inspection robot.

[0194] Solution: Single base station dual signal:

[0195] The two antennas of each base station only transmit navigation signals of one frequency.

[0196] Solution B: Single base station with four signals:

[0197] The two antennas of each base station transmit signals of two frequencies respectively;

[0198] The distribution of GDOP, HDOP and VDOP in the rectangular plane area with coordinate areas of (0,0,3) and (90,145,3) are simulated respectively. The simulation results are shown in Figure 4 , Figure 5 and Figure 6 As shown;

[0199] The defects of the base station layout plan lead to poor DOP distribution effect. Therefore, data sampling in this area should be avoided as much as possible during the experiment. By comparison, it can be found that the distribution effect of VDOP in the positioning area is better than that of HDOP. This is because the base station is located at a higher position and has a better height relative to the receiver.

[0200] Perform a global observation epoch simulation:

[0201] All observation data of 1000 observation epochs are used as a solution unit for positioning. The absolute value of the difference between the coordinates and the actual coordinates in the three directions of East, North, and Up is calculated. The statistical results are as follows: Figure 7 As shown;

[0202] The positioning accuracy of the receiver can be improved by using more epoch observation data for calculation. This section uses 1200 epoch observation data for solution, which can even reach the millimeter level, but it requires the cost of long-term observation. Compared with Scheme A, Scheme B has more observation data in the East, North, and Up directions, higher positioning accuracy, and lower RMSE.

[0203] Comparison with other methods

[0204] The overall error of the method in three directions is statistically analyzed, and the statistical results are as follows: Figure 8As shown, the overall error of the MLP-RL-RF method is smaller than that of the other four methods, which proves the superiority of the MLP-RL-RF method;

[0205] Since the statistical method of error is to use the difference between the calculated result and the actual value, the error obtained can be positive or negative. The error of the MLP-RL-RF method falls relatively evenly around 0, so the average error is very close to 0.

[0206] Example 2

[0207] The robot equipped with the method of the present invention is tested to test the movement ability of the robot.

[0208] Robot speed experiment:

[0209] Based on this method, a ground-based pseudo-satellite navigation and positioning system is generated. The basic speed of the inspection robot refers to the running speed maintained by the inspection robot when it moves along the path planned by the system without instructions from the navigation system. After the navigation system issues an instruction to correct the route, the speed of the robot will change accordingly. The speed of the inspection robot equipped with this system is adjustable, and the speed range is 0.4m / s to 1.0m / s. When constructing a map, the inspection robot's minimum travel speed is 0.3m / s. At this speed, the clarity of the map is guaranteed.

[0210] During the inspection process, the robot performs inspection work at a speed of 1m / s. At this speed, the robot's braking distance is short and it can basically stop immediately, avoiding a series of adverse effects caused by inertia. The robot's motion speed accuracy under different methods is as follows: Fig. 9 As shown in the figure, when the robot moves in a straight line at the minimum speed, inspection speed and maximum speed, the movement process is stable, and the actual movement speed is equivalent to the expected speed. During the movement process, when the robot moves at the minimum speed and inspection speed, it can maintain straight movement. When it moves at the highest speed, the direction starts to deviate slightly after 25m, and the deviation angle is small. In the actual navigation process, the navigation system can be used to fine-tune the movement direction.

[0211] Robot climbing experiment:

[0212] In an outdoor environment, use roads with different slopes as the experimental site. The robot moves at an inspection speed to test the stability of the robot. Turn on the power to ensure that each component can work normally; turn on the wireless communication module, connect to the host computer, place the robot on the ramp, and perform a climbing test. After the climbing test is completed, disconnect, cut off the power, and stop the test.

[0213] When the robot is traveling on a slope with a large slope, it will have a slight shake, and the speed will drop slightly but remain in a straight line. When facing a larger slope, the robot's speed will drop significantly and the driving route will deviate from the straight line. The stability of the robot's climbing path under the five methods is as follows: Fig.10 shown.

[0214] Path planning and navigation experiment:

[0215] When the inspection robot uses laser radar to build a map model of the working environment, the robot uploads the map to the host computer through the Wi-Fi communication module. The host computer sets the inspection work point for the inspection robot. When the robot enters the visual navigation section, the sensor carried by the robot receives the signal from the IC card and switches the navigation mode. When entering this path, the robot switches to the visual navigation mode. When leaving this path, the navigation mode switches back to laser navigation. During the inspection process, the robot accurately reaches the designated work point.

[0216] In the experiment of system path planning and navigation function, we first plan the starting point and target point of the robot. In the lower left corner of the map, we select a point as the starting point of the inspection robot, and then select a target in the upper right corner of the map. The experimental results of path planning and navigation are as follows: Fig.11 shown.

[0217] The path planning system provides clear and concise directions. During driving, the guide points on the path play a role, and the robot's turning process is smooth, reducing the speed and time loss caused by turning, and improving the robot's inspection efficiency. There is a certain deviation between the set value of the inspection robot at the working point and the actual situation on site. In the navigation mode, the robot's maximum error is 2.3cm, which is less than 3cm, meeting the expected requirements of the inspection robot.

[0218] The experimental results show that the basic speed of the robot is stable within the speed range. The reliability of the system is verified through the robot's climbing experiment, and the system's path planning and navigation capabilities are tested. From the test results, it can be seen that the path planned by the system has a clear direction, and the navigation system guides the robot to reach the target point accurately. The inspection function experiment of the system shows that the robot accurately locates the working point during the inspection process, indicating that the system is feasible and can meet the needs of substation inspection robots.

[0219] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0220] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0221] Those skilled in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0222] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0223] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0224] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0225] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0226] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0227] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0228] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0229] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A motion control optimization method for a substation inspection robot based on deep learning, characterized in that: The following steps are involved: Based on the condition number of the normal matrix N0 in the loss function, the pathological condition of the normal matrix N0 is diagnosed, and the normal matrix is ​​corrected by the Tikhonov regularization method, and the three minimum values ​​of the eigenvalues ​​of the normal matrix are adjusted to obtain the corrected normal matrix. Construct path planning method and reward function optimization method based on the corrected normal matrix, and design neural network input and output based on deep learning; Realize the optimal motion control of substation inspection robot.

2. The motion control optimization method of a substation inspection robot based on deep learning according to claim 1 is characterized in that: In step S1, the condition number based on the normal matrix N0 is as follows: Among them, λ1 is the maximum eigenvalue of matrix N0, λ m is the smallest non-zero eigenvalue of the matrix N0.

3. The motion control optimization method of a substation inspection robot based on deep learning according to claim 2 is characterized in that: Spectral decomposition of the normal matrix N0: Among them, Q is an orthogonal matrix.

4. The motion control optimization method of a substation inspection robot based on deep learning according to claim 1 is characterized in that: In step S1, the ill-conditioned model is corrected by using the Tikhonov regularization method, the linearized ill-conditioned model is subjected to a criterion function, and the ill-conditioned model is solved to obtain a stable solution Z of the ill-posed problem.

5. The deep learning-based substation inspection robot motion control optimization method according to claim 4 is characterized in that: The single-difference carrier phase observation equation is solved by using WLS and Tikhonov regularization. After obtaining the floating-point solution and covariance matrix of the single-difference periodic ambiguity, the LAMBDA algorithm is used to fix the ambiguity and obtain the accurate position information of the receiver.

6. The motion control optimization method of a substation inspection robot based on deep learning according to claim 1 is characterized in that: In step S2, the input-output mapping of the neural network is specifically: The historical actions of the inspection robot and the interaction information between the inspection robot's own sensors and the environment are taken as the input of the neural network, and the angular velocity of the left and right wheels of the inspection robot is taken as the output of the neural network. Through the positive and negative rewards of the collision behavior of the reward function, the inspection robot is induced to reach the goal of crossing the inspection point and avoiding collision.

7. The deep learning-based substation inspection robot motion control optimization method according to claim 6 is characterized in that: The reward function is as follows: When r-0 (t) <d nearcol : When the inspection robot arrives at a new inspection point: r(a t ,s t )=r ste n ste +L str r mov +r new n new Time it takes for the inspection robot to complete the checkpoint traversal: r(a t ,s t )=r ste n ste +L str r mov +2r all When the inspection robot collides with an obstacle: r(a t ,s t )=r ste n ste +L str r mov +r col Among them, r ste is the negative ste reward for the number of steps the robot takes, n ste is the number of steps the robot moves, r mov is the negative reward mov for the robot’s moving distance, L str is the linear distance the robot moves from point A to point B; r nearco is the negative reward when the robot approaches collision nearcol, is the distance between the robot and the nearest obstacle r-0 at time t; d is the distance when the robot approaches the collision point, n new is the number of new detection points reached by the new robot; r new The reward for the robot to reach a new detection point; r mov is the reward for reaching all detection points; r col Negative reward col for collision.

8. The deep learning-based substation inspection robot motion control optimization method according to claim 6 is characterized in that: The first control variable of the neural network architecture is the default value of the data package. The structure of the three hidden layers contains 32 neurons each. The multilayer perceptron neural network is connected by a function F(x). The function F(x) consists of multiple linear functions and nonlinear activation functions. The MLP is expressed as: F(x)=λ n ·s n-1 ·l n-1 ……λ1(x)·S Among them, λ n represents each layer i in the neural network, σ is the activation function, S is the softmax function, and the MLP value n of each neuron i,j for: n i,j =f(∑ j w j ·n i-1,j +b j ) Among them, w j is the connection weight of each layer in the neural network, b j is the bias value of the neuron, f is the activation function, n i-1,j is the output value of the jth neuron in the i-1th layer of the neural network.

9. The motion control optimization method of a substation inspection robot based on deep learning according to claim 1 is characterized in that: In step S2, the reward function r(a t ,s t )as follows: Among them, d r-t (t) is the distance from the rt robot to the target detection point at time t; r new is the reward for the new robot to reach the new detection point; r ste is the negative reward ste for the number of steps the robot takes, n ste is the number of steps the robot moves ste, r mov is the negative reward for the robot moving the mov distance, L str is the linear distance the robot moves from point A to point B; r nearcol is the negative reward when the robot nearcol approaches a collision, d r-o (t) is the distance between the robot and the nearest obstacle r-0 at time t; r all The reward for all to reach all checkpoints; r col is the negative reward when col collision occurs, r out is the negative reward when timeout occurs, d nearcol is the distance between the robot nearcol and the obstacle when it is close to collision, r app is the distance when the app robot approaches the detection point, d win is the distance that the robot reaches the detection point at which it is considered to have won; n new is the number of new detection points reached by the robot.

10. A substation inspection robot motion control optimization system based on deep learning, characterized in that: include: The correction module diagnoses the pathological condition of the normal matrix N0 based on the condition number of the normal matrix N0 in the loss function, corrects the normal matrix through the Tikhonov regularization method, adjusts the three minimum values ​​of the eigenvalues ​​of the normal matrix, and obtains the corrected normal matrix; The control module constructs the path planning method and reward function optimization method based on the corrected normal matrix, and designs the neural network input and output based on deep learning; Realize the optimal motion control of substation inspection robot.

11. A computer-readable storage medium storing one or more programs, characterized in that: The one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform the method of any one of claims 1 to 9.

12. A computing device, characterized in that: include: One or more processors, a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include steps for executing the method according to any one of claims 1 to 9.

Images