New energy inspection robot node positioning method and system

Abstract

The embodiment of the invention provides a new energy inspection robot node positioning method and system, and the method comprises the steps: firstly carrying out the dual confidence degree dynamic evaluation of GPS and SLAM data, and quantifying the respective reliability in real time; and then, inputting the confidence into a fusion state machine decision engine, so that the fusion state machine decision engine can intelligently decide that the system should be in an optimal fusion state such as GPS leading, pure SLAM or GPS recovery transition. Finally, the state decision directly guides the adaptive adjustment of the measurement covariance matrix in the extended Kalman filtering, thereby dynamically adjusting the weights of different sensors at the algorithm level. According to the closed-loop design from state perception to strategy decision and then to parameter self-adaption, positioning impact caused by GPS signal sudden change can be effectively restrained, smooth transition during signal recovery is achieved, and therefore it is guaranteed that a positioning result has high precision, high continuity and high accuracy under various working conditions.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application relates to the technical field of intelligent inspection, in particular to a new energy inspection robot node positioning method and system. BACKGROUND

[0002] With the transformation of global energy structure, the new energy industry represented by photovoltaic and wind power develops rapidly. These new energy stations, such as large photovoltaic power stations and wind power stations, usually have wide coverage, numerous equipment nodes and scattered distribution, which puts forward high requirements for the frequency and quality of operation and inspection. The traditional manual inspection mode is not only low in efficiency and high in cost, but also has safety risks in bad weather or complex terrain. Therefore, the use of new energy inspection robots for autonomous and intelligent inspection has become an inevitable trend in the industry. To realize the autonomous inspection of robots, the prerequisite is to obtain the position and attitude of itself in real time, accurately and continuously, that is, to have a high-precision node positioning system. Only when the positioning is accurate, the robot can autonomously navigate according to the preset path, accurately reach each equipment node to be inspected, and complete the data acquisition and analysis tasks.

[0003] In order to realize the accurate positioning of the robot, the prior art adopts a multi-sensor fusion scheme. Among them, the global positioning system, especially the real-time differential technology, can provide centimeter-level absolute position information in an open and unobstructed environment, and is the main means of outdoor robot positioning. At the same time, the simultaneous localization and mapping (SLAM) technology uses sensors such as laser radar or camera to perceive the surrounding environment, estimates the relative motion of the robot by matching environmental features, and can provide continuous positioning in areas where GPS signals are missing (such as indoors, tunnels or high equipment obstruction areas). However, simply fusing these two technologies in the complex scene of new energy inspection will face serious challenges. Specifically, in the scene where the GPS signal is weak or missing, how to ensure the continuity and smoothness of the positioning, and avoid the jump or drift of the positioning result. The reason for this problem is that: on the one hand, there are often tall power generation equipment, metal supports or buildings in new energy stations, which are easy to cause obstruction, attenuation or multipath interference of GPS signals, resulting in a sharp decline in the quality of GPS positioning data or even complete interruption. If the fusion algorithm still trusts and uses these incorrect GPS data, it will seriously pollute the entire positioning system, causing the robot position to jump dramatically; on the other hand, the SLAM technology itself has inherent cumulative error, and in the long-time or long-distance pure SLAM operation mode, the positioning result will gradually deviate from the true trajectory and produce drift. When the robot reenters the area with good GPS signal from the area where the GPS signal is lost, if the fusion strategy simply fuses the restored GPS data with the already drifted SLAM data, it will again cause the abrupt jump of the positioning result due to the large difference between the two positions, which destroys the continuity and smoothness of the positioning trajectory.

[0004] Therefore, an accurate and more optimized new energy inspection robot node positioning scheme is expected. SUMMARY

[0005] Embodiments of the present application aim to at least solve one of the technical problems existing in the prior art, and provide a new energy inspection robot node positioning method and system.

[0006] In a first aspect, embodiments of the present application provide a new energy inspection robot node positioning method, comprising:

[0007] obtaining GPS raw data, SLAM raw data and IMU raw data;

[0008] extracting raw quality indicators from the GPS raw data and the SLAM raw data to obtain SLAM indicators and GPS indicators, the SLAM indicators including the number of valid feature points and a loop closure flag, and the GPS indicators including an RTK state, an HDOP and a number of satellites;

[0009] dynamically evaluating the SLAM indicators and the GPS indicators based on double confidence to obtain GPS confidence and SLAM confidence;

[0010] inputting the GPS confidence and the SLAM confidence into a fusion state machine decision engine to obtain a new fusion state, the new fusion state being GPS dominant, pure SLAM or GPS recovery transition;

[0011] adjusting a measurement covariance based on the GPS confidence and the new fusion state to obtain an adjusted R matrix;

[0012] based on the adjusted R matrix, performing extended Kalman filter update on the GPS raw data, the SLAM raw data and the IMU raw data to obtain a final fused pose.

[0013] In a second aspect, embodiments of the present application provide a new energy inspection robot node positioning system, comprising:

[0014] a raw data acquisition module configured to acquire GPS raw data, SLAM raw data and IMU raw data;

[0015] a quality indicator extraction module configured to extract raw quality indicators from the GPS raw data and the SLAM raw data to obtain SLAM indicators and GPS indicators, the SLAM indicators including the number of valid feature points and a loop closure flag, and the GPS indicators including an RTK state, an HDOP and a number of satellites;

[0016] a dynamic evaluation module configured to perform double confidence dynamic evaluation on the SLAM index and the GPS index to obtain a GPS confidence and a SLAM confidence;

[0017] a fusion state generation module configured to input the GPS confidence and the SLAM confidence into a fusion state machine decision engine to obtain a new fusion state, the new fusion state being GPS dominant or pure SLAM or GPS recovery transition;

[0018] an R matrix adjustment module configured to perform adaptive measurement covariance adjustment based on the GPS confidence and the new fusion state to obtain an adjusted R matrix;

[0019] a fused pose generation module configured to perform extended Kalman filter update on the GPS raw data, the SLAM raw data and the IMU raw data based on the adjusted R matrix to obtain a final fused pose.

[0020] Compared with the prior art, the new energy inspection robot node positioning method and system provided by the application firstly performs double confidence dynamic evaluation on GPS and SLAM data, and quantifies the reliability of each in real time. Subsequently, the confidence is input into a fusion state machine decision engine, so that the system can intelligently decide the optimal fusion state, such as GPS dominant, pure SLAM or GPS recovery transition. Finally, the state decision directly guides the adaptive adjustment of the measurement covariance matrix in the extended Kalman filter, thereby dynamically adjusting the weights of different sensors at the algorithm level. This closed-loop design from state perception to strategy decision to parameter adaptation can effectively suppress the positioning impact caused by the sudden change of GPS signal and realize smooth transition during signal recovery, thereby ensuring that the positioning result has high precision, high continuity and high accuracy under various working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the specific embodiments of the application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or prior art description. Obviously, the drawings described below are some embodiments of the application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0022] Figure 1 a flowchart of the new energy inspection robot node positioning method according to the embodiment of the application;

[0023] Figure 2 a data flow diagram of the new energy inspection robot node positioning method according to the embodiment of the application;

[0024] Figure 3A flow chart for performing extended Kalman filter update on GPS raw data, SLAM raw data and IMU raw data to obtain a final fused pose based on the adjusted R matrix of the node positioning method for the new energy inspection robot according to the embodiment of the application is shown in FIG. 6.

[0025] Figure 4 A flow chart for performing Kalman gain calculation based on the prior covariance matrix, the adjusted R matrix and the prior state estimation to obtain a Kalman gain matrix of the node positioning method for the new energy inspection robot according to the embodiment of the application is shown in FIG. 7.

[0026] Figure 5 A block diagram of the node positioning system for the new energy inspection robot according to the embodiment of the application is shown in FIG. 8. DETAILED DESCRIPTION

[0027] In order for those skilled in the art to better understand the technical solutions of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0028] Unless otherwise specifically defined, the technical terms or scientific terms used in the embodiments of the present application shall be understood as the usual meanings understood by those skilled in the art in the field of the present application. The use of "including" or "containing" and the like in the embodiments of the present application neither limits the mentioned shapes, numbers, steps, actions, operations, components, elements and / or their groups, nor excludes the presence or addition of one or more other different shapes, numbers, steps, actions, operations, components, elements and / or their groups, or addition of these.

[0029] Unless otherwise specifically defined, the relative arrangement of the components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present application. At the same time, it should be understood that, for the convenience of description, the sizes of the various parts shown in the drawings are not drawn in accordance with the actual proportional relationship, and the technology, method and equipment known to those skilled in the related art can not be discussed in detail, but in appropriate cases, the shown technology, method and equipment should be regarded as part of the authorized specification. In all the examples shown and discussed here, any specific other example can have different values. It should be noted that: similar symbols and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0030] In the description of the embodiments of the present invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In the embodiments of the present invention, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in the embodiments of the present invention, as well as the features of different embodiments or examples.

[0031] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0032] This invention proposes a node positioning scheme for a new energy inspection robot to address issues such as positioning result jumps or drifts caused by dynamic changes in GPS signal quality in complex environments, as seen in existing technologies. Specifically, the system first acquires raw data from GPS, SLAM, and IMU in real time, extracting key quality indicators such as RTK status, satellite count, and effective feature point count. These indicators are then quantified into dynamically changing GPS and SLAM confidence scores. These confidence scores are subsequently fed into a core fusion state machine decision engine, which intelligently identifies and switches the system to different operating states, such as GPS-dominated, pure SLAM, or the crucial GPS recovery transition. This macro-state decision directly guides the adaptive adjustment of the core parameter of the extended Kalman filter—the measurement covariance R matrix. In particular, during the GPS recovery transition from interruption to restoration, the system does not abruptly switch back to GPS data. Instead, it gradually integrates the weight of GPS data from low to high within a preset time by smoothly attenuating the covariance, effectively suppressing positioning result jumps caused by sudden changes in data sources. Finally, based on this dynamically adjusted R matrix, deep fusion of multi-source data is achieved through extended Kalman filtering, thereby outputting a final fused pose that is continuous, smooth, and highly accurate under any operating condition, fundamentally solving the problem of unstable positioning during signal switching in traditional fusion methods.

[0033] The present invention proposes a node positioning method for a new energy inspection robot. Figure 1 This is a flowchart of a node positioning method for a new energy inspection robot according to an embodiment of the present invention. Figure 2This is a schematic diagram of the data flow in the node positioning method for a new energy inspection robot according to an embodiment of the present invention. Figure 1 and Figure 2 As shown, the node localization method for a new energy inspection robot according to an embodiment of the present invention includes the following steps: S100, acquiring raw GPS data, raw SLAM data, and raw IMU data; S200, extracting raw quality indicators from the raw GPS data and raw SLAM data to obtain SLAM indicators and GPS indicators, wherein the SLAM indicators include the number of effective feature points and the closed-loop flag, and the GPS indicators include RTK status, HDOP, and the number of satellites; S300, performing dual-reset confidence dynamic evaluation on the SLAM indicators and GPS indicators to obtain GPS confidence and SLAM confidence; S400, inputting the GPS confidence and SLAM confidence into the fusion state machine decision engine to obtain a new fusion state, wherein the new fusion state is GPS-dominated, pure SLAM, or GPS recovery transition; S500, performing adaptive measurement covariance adjustment based on the GPS confidence and the new fusion state to obtain an adjusted R matrix; S600, based on the adjusted R matrix, performing extended Kalman filtering update on the raw GPS data, raw SLAM data, and raw IMU data to obtain the final fused pose.

[0034] Specifically, in step S100, raw GPS data, raw SLAM data, and raw IMU data are acquired. It should be understood that the operating environment of the new energy inspection robot is complex and variable when performing tasks, and a single sensor cannot provide continuous and reliable positioning information. For example, GPS signals may be interrupted or degraded in areas obstructed by large equipment or buildings, leading to positioning failure; while SLAM technology can work in environments without GPS, it has inherent cumulative errors, which can cause positioning drift over long periods; and IMU can provide high-frequency attitude and acceleration information, but integration errors accumulate rapidly. Therefore, in the technical solution of this invention, raw GPS data, raw SLAM data, and raw IMU data are acquired to provide the positioning system with multi-source heterogeneous information with complementary characteristics, namely, the global absolute position of GPS, the local relative position of SLAM, and the high-frequency motion state of IMU. This provides a comprehensive and redundant data foundation for subsequent dynamic evaluation and adaptive fusion algorithms, thereby creating conditions for overcoming the limitations of a single sensor and achieving high-precision positioning in all scenarios.

[0035] More specifically, in a concrete example of the present invention, corresponding sensors are integrated into the hardware platform of the new energy inspection robot, and data acquisition and preprocessing are performed through an onboard computing unit. The data acquisition process of a robot platform for photovoltaic power station inspection is implemented as follows: First, an inertial measurement unit (IMU) is rigidly installed at the geometric center of the robot body. This IMU continuously outputs raw data streams containing three-axis angular velocities and three-axis linear accelerations at a frequency of 100Hz. Second, an RTK-GPS receiver is installed at an unobstructed position on the top of the robot. This receiver analyzes satellite signals and differential base station signals at a frequency of 10Hz, outputting GPS data frames containing latitude and longitude, elevation, RTK status, horizontal accuracy factor (HDOP), and the number of visible satellites. Finally, a three-dimensional lidar is installed at the front end of the robot. This lidar scans the surrounding environment in real time, such as photovoltaic panel arrays and supports, acquiring three-dimensional point cloud data. Simultaneously, the SLAM algorithm module in the onboard computing unit processes this point cloud data, outputting the robot's pose estimate relative to the environment map at a frequency of 10Hz, along with SLAM quality information such as the number of effective feature points and loop closure markers. All sensor data is precisely timestamped for subsequent synchronous fusion processing.

[0036] Specifically, in step S200, raw quality indicators are extracted from the raw GPS data and raw SLAM data to obtain SLAM indicators and GPS indicators. The SLAM indicators include the number of effective feature points and the loop closure flag, and the GPS indicators include RTK status, HDOP, and the number of satellites. It should be understood that the inherent quality and reliability of GPS and SLAM positioning data directly obtained from sensors dynamically change with the environment, and unfiltered direct fusion will introduce unreliable data, thereby polluting the output of the entire positioning system. Therefore, in the technical solution of this invention, raw quality indicators are further extracted from the raw GPS data and raw SLAM data to obtain SLAM indicators and GPS indicators. The SLAM indicators include the number of effective feature points and the loop closure flag, and the GPS indicators include RTK status, HDOP, and the number of satellites, thereby transforming abstract data reliability into quantifiable and specific evaluation indicators. This provides an objective and real-time basis for subsequent confidence assessment and fusion decisions, enabling the system to perceive the quality of its own positioning status.

[0037] More specifically, in a specific example of the present invention, preset key performance parameters are extracted by parsing the data protocol or internal state of the algorithm module output by the sensor. When the inspection robot travels near the tall wind turbine towers in a wind farm, its data quality index extraction process is as follows: For GPS data, the parsing program of the onboard computing unit performs real-time parsing of the NMEA-0183 standard data stream received from the RTK-GPS module, extracts the RTK solution status (such as Fixed, Float, or Single) from the GGA message, and extracts the horizontal accuracy factor HDOP value and the number of satellites used for positioning from the GSA message. For SLAM data, the laser SLAM algorithm running on the computing unit counts the number of laser point cloud features that successfully match the map in the current frame after each pose update and outputs it as the number of valid feature points; at the same time, when the SLAM algorithm identifies that the current position overlaps with a point in the historical trajectory and successfully performs closed-loop optimization, it sets a closed-loop flag. The extracted RTK status, HDOP, number of satellites, number of effective feature points, and closed-loop flags together constitute a set of structured indicators for subsequent evaluation.

[0038] Specifically, in step S300, a dual-reset confidence dynamic evaluation is performed on the SLAM and GPS indicators to obtain the GPS confidence score and SLAM confidence score. It should be understood that the original quality indicators of GPS and SLAM extracted in the previous step have different dimensions and different physical meanings. For example, RTK status is an enumeration type, while the number of satellites is an integer value. These indicators cannot be directly and uniformly quantified for comparison, resulting in the fusion system lacking a comprehensive and intuitive evaluation basis when making decisions. Therefore, in the technical solution of this invention, a dual-reset confidence dynamic evaluation is further performed on the SLAM and GPS indicators to obtain the GPS confidence score and SLAM confidence score, thereby integrating the multi-dimensional, discrete quality indicators into a standardized, continuous confidence score ranging from 0 to 1. In this way, the fusion system can clearly quantify and compare the instantaneous positioning quality of GPS and SLAM at any time. For example, when the inspection robot moves between photovoltaic arrays and causes GPS satellite signals to be blocked, its GPS confidence score will decrease accordingly, while the SLAM confidence score may remain high. The system then obtains a clear basis for making intelligent decisions, providing reliable input for subsequent fusion state switching and weight adaptive adjustment.

[0039] More specifically, in this embodiment of the invention, a dual-reset confidence dynamic evaluation is performed on SLAM and GPS metrics to obtain GPS confidence and SLAM confidence, including: determining RTK score, HDOP score, and satellite number score based on RTK status, HDOP, and satellite number; calculating a weighted sum of RTK score, HDOP score, and satellite number score to obtain the GPS confidence; determining effective feature score and loop closure score based on the number of effective feature points and loop closure flag; and calculating the mean of the effective feature score and loop closure score as the SLAM confidence.

[0040] Specifically, based on RTK status, HDOP, and satellite count, RTK score, HDOP score, and satellite count score are determined; a weighted sum of the RTK score, HDOP score, and satellite count score is calculated to obtain the GPS confidence score. It should be understood that the three key indicators—RTK status, HDOP, and satellite count—obtained from the GPS receiver each reflect positioning quality from different dimensions, but they lack a unified metric, making it impossible to form a comprehensive evaluation conclusion to guide the fusion algorithm. Therefore, in the technical solution of this invention, RTK score, HDOP score, and satellite count score are further determined based on RTK status, HDOP, and satellite count, and their weighted sum is calculated to obtain the GPS confidence score. This normalizes and fuses multiple heterogeneous quality indicators into a single, standardized confidence score value. This allows the system to have an intuitive and quantitative assessment of the overall reliability of the current GPS data. For example, when the inspection robot is operating in an open area, the GPS confidence level calculated by the system will be close to 1.0, indicating that the GPS data is highly reliable. However, when it enters a covered charging shed, the confidence level will drop significantly, thus providing clear quantitative input for subsequent fusion decisions.

[0041] Specifically, in a specific example of the present invention, when the RTK state is Fixed, the RTK score is 1.0; when the RTK state is Float, the RTK score is 0.6; when the RTK state is Single, the RTK score is 0.3; and when the RTK state is None, the RTK score is 0. For an inspection robot performing tasks within a new energy power station, a clear scoring rule and weighted model are established to map various raw indicators into scores and perform comprehensive calculations. First, the discrete RTK states are scored: when the RTK state is Fixed, the RTK score is 1.0; when the RTK state is Float, the RTK score is 0.6; when the RTK state is Single, the RTK score is 0.3; and when the RTK state is None, the RTK score is 0. Secondly, a piecewise or continuous function mapping is applied to the continuous HDOP values ​​and satellite data to obtain scores. For example, when HDOP is less than 1.5, the HDOP score is 1.0; when HDOP is greater than 4.0, the score is 0, with linear interpolation in the intermediate interval. Similarly, when the number of satellites is greater than or equal to 10, the satellite number score is 1.0; when the number of satellites is less than 5, the score is 0, with linear interpolation in the intermediate interval. Finally, these three scores are weighted and summed. For example, if the weighting coefficients are set to 0.5, 0.3, and 0.2, the final GPS confidence score is calculated using the formula: Confidence Score = 0.5 × RTK Score + 0.3 × HDOP Score + 0.2 × Satellite Number Score.

[0042] Specifically, based on the number of effective feature points and the loop closure flag, the effective feature score and the loop closure score are determined; the mean of the effective feature score and the loop closure score is calculated as the SLAM confidence score. It should be understood that the two indicators output from the SLAM system—the number of effective feature points and the loop closure flag—reflect different aspects: one reflects the richness of environmental texture and the quality of immediate matching, while the other reflects the consistency of the global path and the ability to eliminate errors. These two indicators have different properties and require a comprehensive framework to uniformly evaluate their contribution to the overall localization result. Therefore, in the technical solution of this invention, the effective feature score and the loop closure score are further determined based on the number of effective feature points and the loop closure flag, and their mean is calculated as the SLAM confidence score. This integrates SLAM performance indicators reflecting different aspects into a single, standardized confidence score. This allows the fusion system to obtain a quantitative assessment of the current reliability of the SLAM localization scheme. For example, when an inspection robot travels in a long, straight passage with repetitive structures and sparse features (such as a maintenance passage between photovoltaic arrays), the SLAM confidence calculated by the system will naturally decrease, thus providing a basis for fusion decision-making to avoid over-reliance on SLAM results that may drift.

[0043] More specifically, in a concrete example of this invention, when the inspection robot performs an inspection task in the substation area of ​​a new energy power station, the area has a complex structure and rich features. The calculation process of its SLAM confidence score is as follows: First, the number of effective feature points is scored, and a threshold is set. For example, when the number of effective feature points matched by the SLAM algorithm exceeds 100, the environmental features are considered sufficient, and the effective feature score is 1.0; if it is less than 20, the environmental features are considered sparse, and the positioning may be unreliable, with a score of 0.1; if it is between 20 and 100, linear interpolation is used to calculate the score. Second, the loop closure flag is scored. When the SLAM algorithm successfully detects a loop and sets the loop closure flag to 1, it indicates that the global error has been eliminated and the positioning accuracy has been improved, at which point the loop closure score is 1.0; when no loop closure occurs, this score is a neutral benchmark value, such as 0.5. Finally, the arithmetic mean of the two scores is calculated as the final SLAM confidence score. For example, if the current number of valid feature points is 150 (score of 1.0) and a loop closure has just been completed (score of 1.0), then the SLAM confidence score is (1.0 + 1.0) / 2 = 1.0, indicating that SLAM localization is extremely reliable at this time.

[0044] Specifically, in step S400, the GPS confidence level and SLAM confidence level are input into the fusion state machine decision engine to obtain a new fusion state, which is either GPS-dominated, pure SLAM, or a GPS recovery transition. It should be understood that when the new energy inspection robot moves between different work areas, the reliability of GPS and SLAM undergoes a fundamental and non-linear change. Simple linear weighted fusion strategies cannot cope with this qualitative change, especially at the critical moment when the GPS signal is completely lost and then re-acquired. Improper handling can lead to drastic jumps in positioning results. Therefore, in the technical solution of this invention, the GPS confidence level and SLAM confidence level are further input into the fusion state machine decision engine to obtain a new fusion state, which is either GPS-dominated, pure SLAM, or a GPS recovery transition. This transforms continuous confidence level changes into discrete system operating modes with clear physical meaning, thereby achieving macroscopic and logical management of the fusion strategy. This allows the positioning system to smoothly and logically switch between different fusion algorithms or parameter configurations based on the most reliable data source. For example, when the inspection robot enters the indoor charging room, the system can decisively switch to pure SLAM mode to avoid interference from invalid indoor GPS data. More importantly, when it leaves the charging room, the system can enter a dedicated GPS recovery transition state to prepare for subsequent smooth fusion, thereby fundamentally ensuring the continuity and smoothness of the positioning trajectory.

[0045] More specifically, in a specific example of the present invention, GPS confidence and SLAM confidence are input into the fusion state machine decision engine to obtain a new fusion state, which is either GPS-dominated, pure SLAM, or a GPS recovery transition. This includes: switching the new fusion state to pure SLAM in response to the GPS confidence being below 0.2 for one consecutive second; switching the new fusion state from pure SLAM to a GPS recovery transition in response to the GPS confidence recovering from 0.2 to above 0.4; and switching the new fusion state from GPS recovery transition to GPS-dominated in response to the duration of the GPS recovery transition exceeding 3 seconds. A finite state machine (FSM) is implemented in the onboard computing unit, which maintains a current state variable and updates this variable according to preset rules and timers. When the inspection robot moves from an open area to a row of tall inverter equipment rooms within a photovoltaic power station, initially, when the robot is in the open area, the GPS confidence is consistently above 0.4, and the system is in a GPS-dominated state. As the robot approaches and enters the shaded area of ​​the equipment room, the GPS signal quality drops sharply, and the calculated GPS confidence level falls below 0.2. A timer within the state machine starts counting. When the GPS confidence level remains below 0.2 for one consecutive second, the state machine switches the current state from GPS-dominated to pure SLAM. Subsequently, as the robot leaves the shaded area of ​​the equipment room, the GPS signal recovers, the confidence level rises again and exceeds the threshold of 0.4, and the state machine immediately switches the state from pure SLAM to GPS recovery transition, initiating a 3-second transition timer. During this 3-second GPS recovery transition, the system executes a special fusion strategy. When the transition timer reaches 3 seconds, if the GPS confidence level remains stable at a high level, the state machine finally switches the state back from GPS recovery to GPS-dominated, completing a smooth fusion mode switching cycle.

[0046] Specifically, in step S500, adaptive measurement covariance adjustment is performed based on GPS confidence and the new fusion state to obtain the adjusted R matrix. It should be understood that the state machine decisions obtained in the preceding steps (such as GPS-dominated, pure SLAM) are macroscopic-level logical guidance, while the underlying extended Kalman filter algorithm quantifies the uncertainty of measurement data by measuring the covariance R matrix. A precise mathematical bridge is needed between the two to transmit information. If the R matrix is ​​a fixed value, it cannot respond to dynamic changes in sensor quality, resulting in poor fusion performance. Therefore, in the technical solution of this invention, adaptive measurement covariance adjustment is further performed based on GPS confidence and the new fusion state to obtain the adjusted R matrix. This transforms the upper-level state decisions and quantified confidence into a specific and dynamic mathematical description of GPS measurement noise in the Kalman filter. This allows the fusion algorithm to accurately know the extent to which the current GPS measurement value should be trusted in each iteration, thereby achieving precise control from macroscopic strategy to microscopic algorithm parameters, ensuring the adaptability and optimality of the fusion result.

[0047] More specifically, in this embodiment of the invention, adaptive measurement covariance adjustment based on GPS confidence and a new fusion state to obtain an adjusted R matrix includes: if the new fusion state is GPS-dominated, the diagonal elements of the adjusted R matrix are proportional to the reciprocal of the GPS confidence; if the new fusion state is pure SLAM, the values ​​of each element in the adjusted R matrix are set to infinity; if the new fusion state is a GPS recovery transition, a target value of the adjusted R matrix is ​​calculated based on the GPS confidence, and the initial value of the adjusted R matrix is ​​decayed to the target value within a predetermined time. It is worth noting that the predetermined time is 3 seconds. The initial value of the adjusted R matrix is ​​decayed to the target value within the predetermined time using the following formula:

[0048] R(t)=R target +(R initial -R target )*e -k*t

[0049] Where t is time, k is the decay adjustment factor, and R initial As the initial value, R target This is the target value.

[0050] More specifically, in a concrete example of the present invention, when the inspection robot performs a task within a wind farm, the calculation process of its adjusted R matrix is ​​as follows: GPS-dominated state: When the robot travels in an open area with large spacing between wind turbines, the system is in a GPS-dominated state. At this time, the diagonal elements of the adjusted R matrix are set to be proportional to the reciprocal of the current GPS confidence level. For example, if the GPS confidence level is 0.9, the diagonal elements of the R matrix will be set to be smaller accordingly, indicating that the GPS measurement noise is low and the confidence level is high, and its weight in the fusion will be increased accordingly. Pure SLAM state: When the robot enters a large maintenance workshop (where the GPS signal is completely lost), the state machine switches to a pure SLAM state. At this time, all elements of the adjusted R matrix are set to a maximum value within the machine representation range, i.e., equivalent to infinity. Mathematically, this means that the GPS measurement noise is infinitely large, and its weight in the Karl MAN gain calculation will approach zero, thus completely shielding invalid GPS data input at the algorithm level. GPS Recovery Transition State: When the robot leaves the maintenance workshop and receives GPS signals again, the state machine switches to the GPS recovery transition state. At this time, the system first calculates a target R matrix value R based on the currently recovered GPS confidence level. target This value reflects the quality of the GPS data after stable recovery. Meanwhile, the initial value of the R matrix is ​​R0. ihitial It is set to a maximum value (similar to the pure SLAM state). During the next 3-second predetermined time, the R matrix is ​​calculated using the above formula. Through this formula, the value of the R matrix smoothly and exponentially decays from a maximum value to the target value, achieving a soft introduction of GPS data weights starting from zero, effectively avoiding jumps in positioning results caused by sudden intervention of GPS data.

[0051] Specifically, in step S600, based on the adjusted R matrix, the raw GPS data, raw SLAM data, and raw IMU data are updated using extended Kalman filtering to obtain the final fused pose. It should be understood that all preceding steps, including data acquisition, quality assessment, state decision-making, and covariance adjustment, ultimately serve a common goal: how to effectively fuse multi-source heterogeneous sensor information to generate a unified and accurate positioning result. Extended Kalman filtering is a classic and effective mathematical framework for achieving this nonlinear multi-sensor data fusion. Therefore, in the technical solution of this invention, based on the adjusted R matrix, the raw GPS data, raw SLAM data, and raw IMU data are further updated using extended Kalman filtering to obtain the final fused pose. This executes the final calculation and fusion steps of the entire positioning scheme, applying all preprocessing and decision-making results to the actual filtering iteration. In this way, the high-frequency motion prediction provided by the IMU and the dynamically weighted GPS measurements can be optimally combined, thereby outputting a high-precision and smooth continuous final pose at each moment that integrates the advantages of all information sources and effectively suppresses the influence of unreliable data.

[0052] Figure 3 This is a flowchart illustrating the node localization method for a new energy inspection robot according to an embodiment of the present invention, which involves updating the original GPS data, original SLAM data, and original IMU data using extended Kalman filtering based on the adjusted R matrix to obtain the final fused pose. Figure 3 As shown, step S600 includes: S610, performing state vector propagation and uncertainty propagation based on the previous time-state estimate, the previous time-covariance, and the raw IMU data to obtain the prior state estimate and the prior covariance matrix; S620, performing residual calculation on the prior state estimate based on the raw GPS data to obtain the innovation vector; S630, performing Kalman gain calculation based on the prior covariance matrix, the adjusted R matrix, and the prior state estimate to obtain the Kalman gain matrix; S640, performing state and covariance update on the prior state estimate and the prior covariance matrix based on the Kalman gain matrix and the innovation vector to obtain the updated state estimate as the final fused pose.

[0053] Specifically, in step S610, state vector propagation and uncertainty propagation are performed based on the previous time-state estimate, the previous time-covariance, and the raw IMU data to obtain the prior state estimate and the prior covariance matrix. It should be understood that the update frequency of external observation sensors such as GPS and SLAM is relatively low (e.g., 10Hz). Between two effective observations, the robot system needs a high-frequency method to infer its own motion state to ensure the continuity and real-time performance of the positioning output. Therefore, in the technical solution of this invention, state vector propagation and uncertainty propagation are further performed based on the previous time-state estimate, the previous time-covariance, and the raw IMU data to obtain the prior state estimate and the prior covariance matrix. This utilizes the high-frequency (e.g., 100Hz) angular velocity and linear acceleration information provided by the IMU to continuously infer the robot's pose during the Kalman filter prediction stage. This fills the gaps between low-frequency observation data, provides a high-frequency, smooth motion trajectory prediction, and simultaneously quantifies the uncertainty accumulated due to integration, providing a reliable prior benchmark for subsequent observation update steps.

[0054] More specifically, in a concrete example of the present invention, for a robot inspecting between photovoltaic panel arrays, the state prediction process between two GPS updates is as follows: At time k-1, the system has a definite posterior state estimate (including 3D position, 3D velocity, and attitude quaternions) and a corresponding posterior covariance matrix. During the time interval from time k-1 to time k (e.g., 10 milliseconds), the system receives multiple IMU data. For each IMU data, the system substitutes its measured angular velocity and linear acceleration (after deducting the influence of gravity) into a preset nonlinear kinematic equation. By performing discrete-time integration on these equations, the system can calculate the increment of the state vector from time k-1 to the current small time step and accumulate it to the state at time k-1, thereby obtaining a predicted, i.e., prior state estimate. At the same time, the system also calculates the propagation of uncertainty based on the IMU's own noise covariance (a pre-calibrated parameter) and the Jacobian matrix of the state transition equation, i.e., updates the covariance matrix to obtain the prior covariance matrix. This process will be performed continuously at the IMU frequency (e.g., 100Hz) until the next GPS observation time k arrives.

[0055] Specifically, in step S620, residual calculation is performed on the prior state estimate based on the raw GPS data to obtain the innovation vector. It should be understood that the prediction step of the extended Kalman filter is based on the internal calculations of the IMU, and the results accumulate errors over time. GPS, on the other hand, provides external absolute measurements of the system's true state, resulting in a discrepancy between the two. Therefore, in the technical solution of this invention, residual calculation is further performed on the prior state estimate based on the raw GPS data to obtain the innovation vector, thereby quantifying the inconsistency between the system's internal predictions and external actual observations. This allows this discrepancy to be expressed in a clear mathematical vector (i.e., the innovation vector), which directly reflects the direction and magnitude of the predicted state correction, providing the most direct driving force for subsequent Kalman gain calculations and state updates.

[0056] More specifically, in a concrete example of the present invention, when the inspection robot receives a new frame of GPS data in an open area of ​​a new energy power station, the calculation process of its innovation vector is as follows: First, the system extracts the predicted position vector from the prior state estimate of the filter. Since the predicted position is usually based on the robot's own coordinate system or the local map coordinate system, while the original GPS data provides latitude and longitude coordinates, it is necessary to use a coordinate transformation function (e.g., converting latitude, longitude, and altitude to the local ENU coordinate system) to transform the GPS measurement value to the same coordinate system as the prior state estimate. Then, the transformed GPS position vector is directly subtracted from the position vector in the prior state estimate, and the resulting difference vector is the innovation vector. For example, if the predicted position is [100.5, 200.3, 1.1], and the transformed GPS measurement position is [100.2, 200.8, 1.0], then the calculated innovation vector is [-0.3, 0.5, -0.1]. This vector accurately indicates the difference between the predicted value and the actual measured value in the three axes.

[0057] Specifically, in step S630, Kalman gain is calculated based on the prior covariance matrix, the adjusted R matrix, and the prior state estimate to obtain the Kalman gain matrix. It should be understood that the innovation vector calculated in the preceding steps (i.e., the deviation between prediction and measurement) cannot be directly used to correct the state. The system needs an optimal weight to determine the extent to which this deviation is adopted. This weight must simultaneously consider the uncertainty of the prediction itself (represented by the prior covariance matrix) and the uncertainty of the measurement data (represented by the adjusted R matrix). Therefore, in the technical solution of this invention, Kalman gain is further calculated based on the prior covariance matrix, the adjusted R matrix, and the prior state estimate to obtain the Kalman gain matrix, thereby dynamically and optimally balancing the reliability of both prediction and measurement. This ensures that the correction of the prior state is based on a comprehensive consideration of all current uncertainties. For example, when the adjusted R matrix is ​​large (indicating GPS is unreliable), the calculated Kalman gain will be small, thus suppressing the influence of noise data on the fusion result, and vice versa, ultimately achieving adaptive weighting of the innovation vector.

[0058] Figure 4 This is a flowchart illustrating the Kalman gain matrix calculation based on the prior covariance matrix, adjusted R matrix, and prior state estimation in the node localization method for new energy inspection robots according to an embodiment of the present invention. Figure 4 As shown, step S630 includes: S631, calculating the Jacobian matrix of the measurement model at the prior state estimate; S632, calculating the covariance of the innovation based on the Jacobian matrix, the prior covariance matrix, and the adjusted R matrix; S633, calculating the Kalman gain matrix based on the covariance of the innovation, the prior covariance matrix, and the Jacobian matrix.

[0059] Specifically, in step S631, the Jacobian matrix of the measurement model at the prior state estimate is calculated. It should be understood that the system's state vector (containing multiple dimensions such as position, velocity, and attitude) and the GPS measurement vector (usually containing only position information) are inconsistent in dimensionality and physical meaning, making direct covariance calculation impossible. A mathematical tool is needed to establish a linear relationship between the two. Therefore, in the technical solution of this invention, the Jacobian matrix of the measurement model at the prior state estimate is further calculated to correctly map the uncertainty in the state space (represented by the prior covariance matrix P) to the measurement space. This provides a crucial, structurally correct transformation matrix for subsequent calculations of the innovation covariance and Kalman gain, ensuring that the covariance calculation between different physical quantities is mathematically valid and meaningful.

[0060] More specifically, in a concrete example of this invention, assume that the state vector of the inspection robot is defined as a column vector containing nine elements (3D position, 3D velocity, and 3D attitude angle), while the GPS measurement is a 3D position vector. The measurement model h(x) is a mapping function from the 9D state space to the 3D measurement space, which simply extracts the first three position elements from the 9D state vector. To calculate the Jacobian matrix H of the measurement model at the current prior state estimate, it is necessary to calculate the partial derivative of each output component (i.e., x, y, z position) in h(x) with respect to each input component (position, velocity, attitude) in the state vector x. Since the measurement value is only directly related to the position component in the state vector and is independent of the velocity and attitude components, the result of the derivative is a 3x9 matrix. The left side of this matrix is ​​a 3x3 identity matrix (indicating that the partial derivative of the measurement position with respect to the state position is 1), while the 3x6 part on the right side is all zero (indicating that the partial derivatives of the measurement position with respect to the velocity and attitude are 0). The calculated 3x9 Jacobian matrix H precisely establishes the linear mapping relationship between the high-dimensional state space and the low-dimensional measurement space.

[0061] Specifically, in step S632, the covariance of the innovation is calculated based on the Jacobian matrix, the prior covariance matrix, and the adjusted R matrix. It should be understood that the uncertainty of the innovation vector (the difference between prediction and measurement) itself stems from two aspects: first, the uncertainty of the system's own prediction; and second, the uncertainty of external measurement data. Before calculating the Kalman gain, these two uncertainties must be reasonably synthesized to obtain a comprehensive assessment of the overall reliability of the innovation. Therefore, in the technical solution of this invention, the covariance of the innovation is further calculated based on the Jacobian matrix, the prior covariance matrix, and the adjusted R matrix. This allows the covariance of the prediction error to be added to the covariance of the measurement noise in the measurement space, thereby quantifying the overall uncertainty of the innovation vector. This provides a crucial denominator for the subsequent calculation of the Kalman gain. The magnitude of this denominator directly reflects the reliability of the innovation, ensuring that the gain calculation is based on a comprehensive consideration of all sources of uncertainty.

[0062] More specifically, in a concrete example of the present invention, when the inspection robot is running within a new energy power station, the calculation process of its information covariance S is as follows: First, obtain the three key matrices calculated in the preceding steps: the prior covariance matrix P representing the uncertainty of state prediction, the Jacobian matrix H representing the linearization of the measurement model, and the adjusted R matrix representing the uncertainty of the current GPS measurement. Then, the system executes the formula S = H × P × H T +R. This calculation consists of two parts: the first part is H×P×H. T This operation is performed using the Jacobian matrix H and its transpose H. TThe first part correctly transforms and projects the prediction uncertainty P in the state space into the measurement space, obtaining the uncertainty of the predicted value in the measurement dimension. The second part directly adds the adjusted R matrix, introducing the uncertainty of the measurement itself. The final matrix S is the covariance of the innovation, and the magnitude of its diagonal elements intuitively reflects the variance or uncertainty of the innovation vector in each measurement axis.

[0063] Specifically, in step S633, the Kalman gain matrix is ​​calculated based on the covariance of the innovation, the prior covariance matrix, and the Jacobian matrix. It should be understood that the system has already obtained the uncertainty of the predicted state (prior covariance matrix P) and the overall uncertainty of the innovation vector (covariance S of the innovation). Now, a final, optimal scaling factor is needed to determine what proportion of the innovation should be used to correct the prior state estimate. This scaling factor must be the result of a trade-off between prediction uncertainty and innovation uncertainty. Therefore, in the technical solution of this invention, the Kalman gain matrix is ​​further calculated based on the covariance of the innovation, the prior covariance matrix, and the Jacobian matrix to solve for the optimal weight in the sense of minimum mean square error. This ensures that the magnitude of the state update is adaptive: when the prediction uncertainty is much greater than the measurement uncertainty, the calculated Kalman gain will be larger, and the system will trust the measurement more; conversely, when the measurement uncertainty (reflected in S) is large, the gain will be smaller, and the system will rely more on its own prediction, thus achieving optimal fusion of prediction and measurement.

[0064] More specifically, in a concrete example of this application, after the inspection robot completes a GPS measurement and calculates all the necessary intermediate matrices within the new energy power station, the final calculation process of its Kalman gain matrix K is as follows: The system calls three key matrices: the prior covariance matrix P, the Jacobian matrix H, and the innovation covariance matrix S. Then, the system executes the formula K = P × h T ×S (-1) Logically, this operation can be understood as a ratio calculation: numerator P × H T This represents the projection of the uncertainty in state prediction onto the measurement space, while the denominator S... (-1) This represents the reciprocal of the overall uncertainty of the new information. When the GPS signal quality is high, the adjusted R matrix is ​​small, resulting in a relatively small S, and the inverse of S is S... (-1) If the value of S is large, the calculated Kalman gain K will be large, meaning the system will assign a larger weight to GPS measurements to correct the prediction. Conversely, when the robot enters an area where GPS signals are blocked, the adjusted R matrix is ​​set to a maximum value, causing S to also be maximized. (-1) As the Kalman gain approaches zero, the calculated Kalman gain K also approaches zero, thus almost completely ignoring unreliable GPS measurements in the update step and ensuring the stability of the positioning.

[0065] Specifically, in step S640, based on the Kalman gain matrix and the innovation vector, the prior state estimate and the prior covariance matrix are updated to obtain the updated state estimate as the final fused pose. It should be understood that the prediction stage of the extended Kalman filter only provides a preliminary estimate based on internal calculations, while the innovation vector and Kalman gain calculated in the preceding steps contain correction information and optimal weights from external measurements. A final synthesis step is needed to fuse this information to obtain a more accurate result than simple prediction or simple measurement. Therefore, in the technical solution of this invention, the prior state estimate and the prior covariance matrix are further updated based on the Kalman gain matrix and the innovation vector to obtain the updated state estimate as the final fused pose, thereby completing the last step of the Kalman filter loop, i.e., performing correction of the predicted state. In this way, the high-frequency predictions of the IMU can be combined with the optimally weighted GPS observation corrections, which not only generates the most reliable robot pose at the current moment, but also updates the uncertainty description of the system itself, providing a more accurate starting point for the prediction of the next filtering cycle, thereby realizing the closed loop and iterative optimization of the entire fusion localization process.

[0066] More specifically, in a concrete example of this invention, after the inspection robot completes the Kalman gain calculation within the new energy power station, the final pose generation process is as follows: State Update: The system acquires the prior state estimation vector obtained in the prediction phase, as well as the Kalman gain matrix K and innovation vector calculated in the previous steps. Then, it executes the state update formula: Posterior State Estimation = Prior State Estimation + K × Innovation Vector. The physical meaning of this operation is to add a correction amount weighted by the Kalman gain to the prior estimate. For example, when the GPS signal is good and the Kalman gain K is large, the innovation vector will be adopted to a greater extent, making the updated state closer to the GPS measurement value. Conversely, when the GPS signal is poor and K approaches zero, the correction amount also approaches zero, and the updated state will basically remain the prior estimate value, that is, mainly relying on the IMU's calculation. Covariance Update: At the same time, the system also needs to update the uncertainty of the state. It acquires the prior covariance matrix P, Kalman gain K, and Jacobian matrix H, and executes the covariance update formula: posterior covariance matrix = (IK × H) × prior covariance matrix, where I is the identity matrix. The significance of this operation is that the overall uncertainty of the system is reduced due to the introduction of external measurement information. The updated posterior covariance matrix P is smaller than the prior value, reflecting an improved confidence level in the fused state estimate. Finally, the posterior state estimate obtained after the above state and covariance updates is used as the final fused pose output for the current moment and as the starting state for the next filtering cycle, thereby achieving continuous and accurate localization.

[0067] In summary, the node localization method for a new energy inspection robot according to embodiments of the present invention is explained. First, it performs a dual-reset confidence dynamic assessment of GPS and SLAM data, quantifying their respective reliability in real time. Then, this confidence level is input into a fusion state machine decision engine, enabling it to intelligently determine the optimal fusion state for the system, such as GPS-dominated, pure SLAM, or GPS recovery transition. Finally, this state decision directly guides the adaptive adjustment of the measurement covariance matrix in the extended Kalman filter, thereby dynamically adjusting the weights of different sensors at the algorithm level. This closed-loop design, from state perception to strategy decision-making to parameter adaptation, effectively suppresses the positioning shock caused by sudden changes in GPS signals and achieves a smooth transition during signal recovery, thus ensuring high precision, high continuity, and high accuracy in positioning results under various operating conditions.

[0068] Furthermore, a node positioning system for a new energy inspection robot is also provided.

[0069] Figure 5 This is a block diagram of a node positioning system for a new energy inspection robot according to an embodiment of the present invention. Figure 5 As shown, the new energy inspection robot node positioning system 500 according to an embodiment of the present invention includes: a raw data acquisition module 510, used to acquire GPS raw data, SLAM raw data, and IMU raw data; a quality index extraction module 520, used to extract raw quality indicators from the GPS raw data and SLAM raw data to obtain SLAM indicators and GPS indicators, wherein the SLAM indicators include the number of effective feature points and the closed-loop flag, and the GPS indicators include RTK status, HDOP, and the number of satellites; and a dynamic evaluation module 530, used to perform dual-reset reliability dynamic evaluation on the SLAM indicators and GPS indicators. The system estimates GPS confidence and SLAM confidence. A fusion state generation module 540 inputs the GPS confidence and SLAM confidence into the fusion state machine decision engine to obtain a new fusion state, which is either GPS-dominated, purely SLAM, or a GPS recovery transition. An R-matrix adjustment module 550 performs adaptive measurement covariance adjustment based on the GPS confidence and the new fusion state to obtain an adjusted R-matrix. A fusion pose generation module 560 updates the original GPS data, original SLAM data, and original IMU data using extended Kalman filtering based on the adjusted R-matrix to obtain the final fusion pose.

[0070] As described above, the new energy inspection robot node positioning system 500 according to embodiments of the present invention can be implemented in various wireless terminals, such as servers with new energy inspection robot node positioning algorithms. In one possible implementation, the new energy inspection robot node positioning system 500 according to embodiments of the present invention can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the new energy inspection robot node positioning system 500 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the new energy inspection robot node positioning system 500 can also be one of many hardware modules of the wireless terminal.

[0071] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A node positioning method for a new energy inspection robot, characterized in that, include: Acquire raw GPS data, raw SLAM data, and raw IMU data; Raw quality indicators are extracted from raw GPS data and raw SLAM data to obtain SLAM indicators and GPS indicators. The SLAM indicators include the number of effective feature points and the loop closure flag. The GPS indicators include RTK status, HDOP and number of satellites. A dual-reset confidence dynamic evaluation was performed on the SLAM and GPS metrics to obtain the GPS confidence and SLAM confidence. The GPS confidence and SLAM confidence are input into the fusion state machine decision engine to obtain a new fusion state, which is either GPS-dominated, pure SLAM, or GPS recovery transition. Adaptive measurement covariance adjustment is performed based on GPS confidence and the new fusion state to obtain the adjusted R matrix; Based on the adjusted R matrix, extended Kalman filtering is applied to the raw GPS data, raw SLAM data, and raw IMU data to obtain the final fused pose.

2. The node positioning method for a new energy inspection robot according to claim 1, characterized in that, A dual-reset confidence dynamic evaluation is performed on SLAM and GPS metrics to obtain GPS confidence and SLAM confidence, including: The RTK score, HDOP score, and satellite number score are determined based on RTK status, HDOP, and satellite number. The weighted sum of the RTK score, HDOP score, and satellite number score is calculated to obtain the GPS confidence level; Based on the number of effective feature points and the closed-loop flag, the effective feature score and the closed-loop score are determined. The mean of the effective feature scores and the loop closure scores is calculated as the SLAM confidence level.

3. The node positioning method for a new energy inspection robot according to claim 2, characterized in that, When the RTK state is Fixed, the RTK score is 1.0; when the RTK state is Float, the RTK score is 0.6; when the RTK state is Single, the RTK score is 0.3; and when the RTK state is None, the RTK score is 0.

4. The node positioning method for a new energy inspection robot according to claim 1, characterized in that, The GPS confidence and SLAM confidence are input into the fusion state machine decision engine to obtain a new fusion state, which is either GPS-dominated, purely SLAM, or a GPS recovery transition, including: If the GPS confidence level falls below 0.2 for one consecutive second, the new fusion state is switched to pure SLAM. In response to the GPS confidence level recovering from 0.2 to above 0.4, the new fusion state is switched from pure SLAM to GPS recovery transition; If the duration of the GPS recovery transition exceeds 3 seconds, the new fusion state is switched from GPS recovery transition to GPS dominance.

5. The node positioning method for a new energy inspection robot according to claim 1, characterized in that, Adaptive measurement covariance adjustment based on GPS confidence and the new fusion state is used to obtain the adjusted R matrix, including: If the new fusion state is GPS-dominated, the diagonal elements of the adjusted R matrix are proportional to the reciprocal of the GPS confidence level; If the new fusion state is pure SLAM, the values ​​of each element in the adjusted R matrix are set to infinity; If the new fusion state is a GPS recovery transition, the target value of the adjusted R matrix is ​​calculated based on the GPS confidence level, and the initial value of the adjusted R matrix is ​​decayed to the target value within a predetermined time.

6. The node positioning method for a new energy inspection robot according to claim 5, characterized in that, The predetermined time is 3 seconds; The process of attenuating the initial value of the adjusted R matrix to the target value within a predetermined time includes: attenuating the initial value of the adjusted R matrix to the target value within a predetermined time using the following formula: R(t)=R target +(R initial -R target )*e -k*t Where t is time, k is the decay adjustment factor, and R initial As the initial value, R target This is the target value.

7. The node positioning method for a new energy inspection robot according to claim 1, characterized in that, Based on the adjusted R matrix, extended Kalman filtering is applied to the raw GPS data, raw SLAM data, and raw IMU data to obtain the final fused pose, including: State vector propagation and uncertainty propagation are performed based on the previous time-state estimate, the previous time-covariance, and the raw IMU data to obtain the prior state estimate and the prior covariance matrix; Based on the raw GPS data, residual calculation is performed on the prior state estimate to obtain the innovation vector; Kalman gain matrix is ​​obtained by calculating the Kalman gain based on the prior covariance matrix, the adjusted R matrix and the prior state estimate. Based on the Kalman gain matrix and the innovation vector, the prior state estimate and the prior covariance matrix are updated to obtain the updated state estimate as the final fused pose.

8. The node positioning method for a new energy inspection robot according to claim 7, characterized in that, The Kalman gain matrix is ​​obtained by calculating the Kalman gain based on the prior covariance matrix, the adjusted R matrix, and the prior state estimate, including: Calculate the Jacobian matrix of the measurement model at the prior state estimate; The covariance of the innovation is calculated based on the Jacobian matrix, the prior covariance matrix, and the adjusted R matrix. The Kalman gain matrix is ​​calculated based on the covariance of the new information, the prior covariance matrix, and the Jacobian matrix.

9. A node positioning system for a new energy inspection robot, characterized in that, include: The raw data acquisition module is used to acquire raw GPS data, raw SLAM data, and raw IMU data. The quality index extraction module is used to extract raw quality indexes from GPS raw data and SLAM raw data to obtain SLAM indexes and GPS indexes. The SLAM indexes include the number of effective feature points and the loop closure flag, and the GPS indexes include RTK status, HDOP and number of satellites. The dynamic evaluation module is used to perform dual-reset confidence dynamic evaluation of SLAM and GPS indicators to obtain GPS confidence and SLAM confidence. The fusion state generation module is used to input GPS confidence and SLAM confidence into the fusion state machine decision engine to obtain a new fusion state, which is a GPS-dominated, pure SLAM, or GPS recovery transition. The R-matrix adjustment module is used to adaptively adjust the measurement covariance based on GPS confidence and the new fusion state to obtain the adjusted R-matrix. The fusion pose generation module is used to update the raw GPS data, raw SLAM data, and raw IMU data with extended Kalman filtering based on the adjusted R matrix to obtain the final fusion pose.

10. The new energy inspection robot node positioning system according to claim 9, characterized in that, The dynamic evaluation module is also specifically used for: The RTK score, HDOP score, and satellite number score are determined based on RTK status, HDOP, and satellite number. The weighted sum of the RTK score, HDOP score, and satellite number score is calculated to obtain the GPS confidence level; Based on the number of effective feature points and the closed-loop flag, the effective feature score and the closed-loop score are determined. The mean of the effective feature scores and the loop closure scores is calculated as the SLAM confidence level.

Images