A high-precision positioning method combining UWB and IMU for complex hydropower plant environments

Abstract

This application relates to the field of spatial positioning technology and discloses a high-precision positioning method integrating UWB and IMU for complex hydropower plant environments. The method is based on a pre-defined electromagnetic isolation mechanism and spatial transition zone to determine the initial node topology and establish a global coordinate system. It analyzes radio frequency signals to extract channel impulse response characteristics, performs frequency domain filtering on the inertial data stream based on characteristic frequency parameters, and extracts vibration intensity markers and pre-integrated datasets. The aforementioned radio frequency features are input into a classification model to identify multipath environments, perform error compensation, and output an observation noise covariance matrix with penalty weights. The vibration intensity markers are converted into a noise covariance matrix for a dynamic adjustment process of the expansion coefficient, and the positioning coordinates are calculated using a Kalman filter based on the aforementioned data input error state. Finally, a dynamic restricted area is constructed using a virtual model to perform over-limit warnings. This effectively overcomes the positioning divergence problem caused by the dual extreme interference of strong magnetic fields and unit resonance, achieving a high-precision, continuous, and safe positioning closed loop.

Description

Technical Field

[0001] This invention relates to the field of spatial positioning technology, and in particular to a high-precision positioning method that integrates UWB and IMU for complex hydropower plant environments. Background Technology

[0002] In the field of industrial IoT and personnel safety positioning, tightly coupled fusion positioning of ultra-wideband (UWB) and inertial measurement unit (IMU) is currently the mainstream technical solution. Chinese patent CN119110244A, "An Ultra-Wideband UWB Real-Time Positioning Method and System for Obstacle-Occlusion Scenarios," discloses a UWB and IMU data fusion architecture based on an error state Kalman filter (ESKF) and a nonlinear optimization algorithm. This technology is mainly aimed at the situation in conventional industrial or warehouse environments where UWB signals suffer from non-line-of-sight errors due to static obstacles. Under normal conditions, it fuses UWB observations and IMU calculations. When the system determines that UWB is occluded, the algorithm logic tends to reduce or isolate the measurement weight of UWB and instead rely entirely on the inertial calculations of the IMU to maintain the continuity of the positioning track.

[0003] However, the aforementioned existing technologies share a common industry assumption: when UWB fails due to obstruction, IMU data serves as a reliable relative benchmark. When facing complex industrial environments with specific extreme physical and electromagnetic interference, such as hydropower plants, this approach neglects the engine frequency resonance pollution of specific industrial equipment, leading to rapid divergence in inertial calculations. When a target approaches the unit, causing severe UWB multipath distortion, the logic of the existing technology will forcibly switch to IMU dominance. However, at this time, the underlying physical structure of the IMU is under high-frequency engine frequency vibration interference, and its accelerometer and gyroscope raw data already contain significant high-frequency noise. If traditional integral calculations are performed directly, the trajectory will diverge exponentially in a short time, causing filter instability. Simultaneously, in complex industrial environments, existing technologies cannot effectively distinguish between the fixed time delay caused by static concrete load-bearing walls and the severe Doppler shift caused by the high-speed rotation of dynamic metal surfaces. Due to the lack of refined analysis of the underlying radio frequency waveform physical characteristics, the absolute isolation or constant weighting strategies employed can easily lead to the false negativeing ​​of effective observation data or the omission of severely contaminated distorted data, causing high-frequency random jumps in the global positioning trajectory. In summary, there is an urgent need for a novel fusion positioning architecture that introduces a mechanism for identifying and pre-filtering specific machine frequency vibrations. At the same time, it is necessary to deeply extract the channel impulse response characteristics of radio frequency signals to accurately distinguish multipath interference attributes, and to implement adaptive adjustment of dynamic process noise and observation penalty weights within the filter, so as to ensure stable, continuous and high-precision three-dimensional positioning coordinates output by the system under dual interference. Summary of the Invention

[0004] The main objective of this invention is to provide a high-precision positioning method that integrates UWB and IMU for complex hydropower plant environments, thereby solving the problems of fusion positioning divergence and trajectory drift caused by the dual interference of strong magnetic fields and machine frequency resonance in hydropower plants.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a high-precision positioning method for UWB and IMU fusion for complex hydropower plant environments, characterized by including the following steps: Based on the preset electromagnetic isolation mechanism and spatial transition zone, determine the initial physical node topology of the positioning object and base station anchor point, establish a global spatial coordinate system and initialize the system's reference state variable set; The raw radio frequency signals collected by the positioning object are analyzed to extract the channel impulse response feature set; at the same time, based on the preset feature frequency parameters, subframe segmentation and frequency domain filtering are performed on the high-frequency raw inertial data stream collected by the positioning object to extract vibration intensity identifiers and generate vibration-resistant pre-integrated inertial datasets. The channel impulse response feature set is input into the non-line-of-sight classification model to identify the current multipath interference environment and generate a non-line-of-sight environment status identifier. When the non-line-of-sight environment status identifier is generated, the corresponding error compensation algorithm is called to calculate and output the final radio frequency observation vector and the penalized observation noise covariance matrix. The vibration intensity label is converted into an expansion coefficient to dynamically adjust the dynamic process noise covariance matrix; the adjusted dynamic process noise covariance matrix is ​​combined with iterative calculation to generate the prediction error state covariance matrix; the prediction error state covariance matrix, the final radio frequency observation vector, the penalized observation noise covariance matrix, and the anti-vibration pre-integration inertial dataset are input into the error state Kalman filter to perform measurement update and combination calculation, and output the final positioning coordinates; The final positioning coordinates are mapped to a preset physical virtual model, and a dynamic restricted area is constructed by combining the physical radiation properties of the physical virtual model. When the positioning coordinates intrude into the dynamic restricted area, an over-limit warning signal and feedback control flow are output for the positioning object based on the spatial intersection calculation result.

[0006] In the preferred scheme, based on the preset electromagnetic isolation mechanism and spatial transition zone, the initial physical node topology of the positioning object and base station anchor point is determined, and a global spatial coordinate system is established, including: The radiation distribution boundary of the alternating magnetic field is determined based on the spatial radiation parameters of the hydro-generator unit, and a high permeability shielding material is set inside the casing of the sensing device carried by the positioning object as an electromagnetic isolation layer. Based on the physical structure diagram of the factory, areas with field strength below the preset safety threshold are selected as safe deployment zones. Strong magnetic radiation sectors are avoided, and base station anchor points are fixed in the safe deployment zones to form the initial physical node state. Using the factory building elevation data, the through area with abrupt elevation change on the Z-axis is extracted and marked as a spatial transition zone. When the base station anchor point falls into the spatial transition zone, the overlap coverage check of the vertical pitch angle is performed based on the elevation data to generate a set of base station anchor points with absolute spatial reference information, and a global spatial coordinate system is established accordingly.

[0007] In the preferred scheme, establishing a global spatial coordinate system and initializing the system's reference state variable set also includes multi-source time coordination and state initialization steps, specifically including: After detecting that the positioning object is in a power-on active state, the precise time protocol is invoked to perform hardware-level time alignment between the inertial data sampling clock and the radio frequency ranging clock of the positioning object, and generate an alignment timestamp; Within the preset static state window, the control positioning object collects the first segment of silent inertial data, calculates and extracts the initial zero bias data of the inertial sensor; The initial spatial coordinates and initial attitude of the positioning object are calculated by using the ranging pulse received from the base station anchor point and combining the global spatial coordinate system. The initial zero-bias data, initial spatial coordinates, and initial attitude are aggregated and encapsulated to generate a set of baseline state variables, which serve as the starting point for dynamic solution iteration.

[0008] In the preferred embodiment, the raw radio frequency signal collected by the positioning object is analyzed to extract the channel impulse response feature set; simultaneously, based on preset feature frequency parameters, subframe segmentation and frequency domain filtering are performed on the high-frequency raw inertial data stream collected by the positioning object to extract vibration intensity identifiers and generate a vibration-resistant pre-integrated inertial dataset, including: In the digital baseband processing stage, the original radio frequency signal is analyzed to separate the peak attenuation slope feature that characterizes the multipath attenuation rate, the first path arrival power feature that characterizes the direct path energy, and the total received power feature that characterizes the overall signal energy. The features are then spliced ​​together to construct a set of channel impulse response features. The rated rotational frequency of the hydro-generator unit is used as the characteristic frequency parameter and the corresponding fundamental harmonic period is calculated. The fundamental harmonic period is used as the time slice threshold to divide the high-frequency raw inertial data stream into discrete inertial subframe sequences. The inertial subframe sequence is filtered using a preset frequency domain filtering operator. When a high-frequency noise energy peak that matches the characteristic frequency parameter and its harmonics is detected, the vibration noise in that frequency band is removed. The energy characteristics of the residual noise in the filtered subframe are calculated and quantized as vibration intensity indicators. At the same time, the denoised subframe data is pre-integrated and accumulated to generate an anti-vibration pre-integrated inertial dataset.

[0009] In the preferred scheme, the channel impulse response feature set is input into the non-line-of-sight classification model to identify the current multipath interference environment and generate a non-line-of-sight environment status identifier. When the non-line-of-sight environment status identifier is generated, the corresponding error compensation algorithm is called to calculate and output the final radio frequency observation vector and the penalized observation noise covariance matrix, including: The channel impulse response feature set is loaded into a preset neural network classification model to calculate the time dimension attenuation gradient of the peak attenuation slope feature, and to calculate the dynamic ratio between the first path arrival power feature and the total received power feature. Based on the periodic jitter characteristics of the ratio between the constant state and dynamic value of the attenuation gradient, the multipath interference environment is determined to be either static occlusion or dynamic metallic occlusion, and the corresponding non-line-of-sight environment status label is generated. When the non-line-of-sight environment status indicator is characterized as static occlusion, a preset nonlinear regression compensation operator is invoked to subtract the penetration time delay in the ranging data to generate a smoothed and corrected ranging value, and the variance is increased simultaneously. When the non-line-of-sight environment status indicator is characterized as dynamic metallic occlusion, a limit observation noise penalty weight is calculated based on the degree of signal distortion to suppress the ranging data, and the final ranging value is encapsulated into a final radio frequency observation vector. The variance and the limit observation noise penalty weight are encapsulated to generate a penalized observation noise covariance matrix.

[0010] In the preferred embodiment, the vibration intensity identifier is converted into an expansion coefficient to dynamically adjust the dynamic process noise covariance matrix; the adjusted dynamic process noise covariance matrix is ​​then used for iterative calculation to generate the prediction error state covariance matrix, including: Obtain the last corrected state variable from the reference state variable set at the previous moment, and use the anti-vibration pre-integrated inertial dataset to perform dynamic integral derivation of the nominal kinematic equation in the last corrected state variable at the previous moment, and calculate and generate the predicted nominal state at the current moment. The vibration intensity label is input into a preset covariance dynamic weighting function for feature mapping to obtain the corresponding expansion coefficient; The corresponding weights in the dynamic process noise covariance matrix are dynamically amplified using an expansion coefficient; combined with the amplified and adjusted dynamic process noise covariance matrix, the error state covariance propagation equation is iteratively derived and calculated to generate the prediction error state covariance matrix.

[0011] In the preferred scheme, after outputting the final RF observation vector and the penalized observation noise covariance matrix, a time-series alignment step for cross-modal data is also included, specifically: Extract the hardware receiving timestamp corresponding to the final RF observation vector, and traverse the buffered anti-vibration pre-integration inertial dataset within the preset timestamp alignment buffer; use a sliding window interpolation algorithm to match and extract the pre-integration data slices aligned with the hardware receiving timestamp in the anti-vibration pre-integration inertial dataset. The final radio frequency observation vectors with the same time base, the penalized observation noise covariance matrix, and the extracted pre-integrated data slices are encapsulated to generate a cross-modal synchronous fusion observation queue.

[0012] In the preferred scheme, the predicted error state covariance matrix, the final RF observation vector, the penalized observation noise covariance matrix, and the vibration-resistant pre-integrated inertial dataset are input into the error state Kalman filter to perform measurement update and combination calculations, outputting the final positioning coordinates, including: Based on the predicted nominal state, a predicted observation vector is generated by mapping the measurement equation, and the difference between the predicted observation vector and the final radio frequency observation vector is calculated to obtain the observation residual vector. By performing matrix inversion operations on the state covariance matrix of the prediction error and the covariance matrix of the penalized observation noise, and combining the adaptive compression effect of the penalty weight, the Kalman gain after bottom-level shielding compression is calculated and generated. The error state vector is generated by multiplying the Kalman gain by the observation residual vector. The error components in the error state vector are compensated to the corresponding nominal components of the predicted nominal state and combined to generate the final corrected state variable. The variable is then substituted into the covariance update equation to calculate the posterior error state covariance matrix. The spatial three-dimensional coordinate matrix is ​​extracted from the final corrected state variable and the final positioning coordinates are obtained by solving the equation.

[0013] In the preferred embodiment, the final positioning coordinates are mapped to a preset virtual entity model, and a dynamic restricted area is constructed based on the physical radiation properties of the virtual entity model. When the positioning coordinates intrude into the dynamic restricted area, an over-limit warning signal and feedback control flow are output for the positioning object based on the spatial intersection calculation result, including: The final positioning coordinates are imported into a preset 3D solid rendering engine, and the factory building elevation data and solid virtual model are analyzed to construct a reference space mapping mesh. By performing geometric matrix affine transformation, the final positioning coordinates are projected and mounted into the mapping mesh, and a 3D trajectory mapping map is generated. Simultaneously analyze the physical radiation properties and the preset safe electronic boundary to construct a dynamic restricted area with magnetic field hazard properties and elevation hazard properties; The real-time updated final positioning coordinate sequence is subjected to spatial Boolean intersection operation with the dynamic restricted area. When the operation result determines that there is coordinate intrusion, an over-limit warning signal is generated through logical branches according to the spatial depth of the intrusion and the level of danger attribute. The over-limit warning signal is packaged into a protocol and a feedback control flow is generated and sent to the positioning object to trigger a physical alarm.

[0014] In a preferred embodiment, the present invention provides a high-precision positioning system fused with UWB and IMU for complex hydropower plant environments, characterized by comprising: The baseline initialization module is used to determine the initial physical node topology of the positioning object and the base station anchor point according to the preset electromagnetic isolation mechanism and spatial transition zone, establish a global spatial coordinate system and initialize the system's baseline state variable set. The anti-interference preprocessing module is used to parse the raw radio frequency signals collected by the positioning object to extract the channel impulse response feature set; at the same time, based on the preset feature frequency parameters, it performs subframe segmentation and frequency domain filtering on the high-frequency raw inertial data stream collected by the positioning object, extracts vibration intensity identifiers and generates anti-vibration pre-integrated inertial dataset. The error compensation and alignment module is used to input the channel impulse response feature set into the non-line-of-sight classification model, identify the current multipath interference environment and generate a non-line-of-sight environment status identifier; when the non-line-of-sight environment status identifier is generated, the corresponding error compensation algorithm is called to calculate and output the final radio frequency observation vector and the penalized observation noise covariance matrix. The tightly coupled solution module is used to convert vibration intensity indicators into expansion coefficients to dynamically adjust the dynamic process noise covariance matrix; it combines the adjusted dynamic process noise covariance matrix to perform iterative calculations to generate the prediction error state covariance matrix; it inputs the prediction error state covariance matrix, the final RF observation vector, the penalized observation noise covariance matrix, and the anti-vibration pre-integration inertial dataset into the error state Kalman filter to perform measurement update and combination calculations, and outputs the final positioning coordinates; The twin mapping and early warning module is used to map the final positioning coordinates to a preset physical virtual model and construct a dynamic restricted area based on the physical radiation properties of the physical virtual model. When the positioning coordinates intrude into the dynamic restricted area, the module outputs an over-limit warning signal and feedback control flow for the positioning object based on the spatial intersection calculation result.

[0015] This invention provides a high-precision positioning method combining UWB and IMU for complex hydropower plant environments. Addressing the unique characteristics of hydropower plants, such as strong magnetic radiation and large elevation differences, this application not only physically cuts off electromagnetic pollution of the positioning system's underlying data by setting up an electromagnetic isolation mechanism and fixing base station anchor points in a safe deployment area, but also overcomes the positioning discontinuity problem caused by cross-floor Z-axis coordinate jumps at the geometric model level by extracting spatial transition zones and performing overlapping coverage checks on base station anchor points. This ensures the continuity of the absolute coordinate system and the underlying stability of the system operation in multi-layered complex spaces. Furthermore, in response to the problem that high-frequency resonance generated by the operation of hydro-generator units can easily lead to rapid divergence of inertial integrals, this solution creatively introduces a subframe frequency domain filtering mechanism based on specific characteristic frequency parameters. By directly removing high-frequency noise energy peaks that match the unit harmonics in the front-end processing stage and extracting vibration intensity indicators, the purity of the inertial estimation benchmark during the flow interruption is enhanced from the source, avoiding the erroneous injection of highly weighted divergent data into the positioning solution system under complex vibration conditions. Furthermore, this scheme deeply extracts the channel impulse response waveform characteristics of the original radio frequency signal, uses a classification model to intelligently identify static and dynamic metal blockage states, and generates a limit observation noise penalty weight when dynamic blockage is determined. In the subsequent filtering and solving stage, the aforementioned vibration intensity label is converted into an expansion coefficient to dynamically adjust the process noise covariance matrix, and combined with the penalty weight, adaptive compression of the Kalman gain is achieved. This enables the system to smoothly and intelligently suppress contaminated observation data and unreliable predicted paths when facing the dual harsh conditions of dynamic multipath on the metal curved surface of a large generator and high-frequency vibration of the foundation, rather than rigidly switching weights between two extreme sources, thereby outputting a stable and continuous high-precision global trajectory. Attached Figure Description

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of a high-precision positioning method using UWB and IMU fusion for complex hydropower plant environments, according to the present invention. Figure 2 This is a schematic diagram of a high-precision positioning system fused with UWB and IMU for complex hydropower plant environments, according to the present invention. Detailed Implementation

[0017] Example 1 like Figure 1-2 As shown, a high-precision positioning method combining UWB and IMU for complex hydropower plant environments is proposed, which includes the following steps: S1. Based on the preset electromagnetic isolation mechanism and spatial transition zone, determine the initial physical node topology of the positioning object and base station anchor point, establish a global spatial coordinate system and initialize the system's reference state variable set. S2. Analyze the raw radio frequency signal collected by the positioning object to extract the channel impulse response feature set; at the same time, perform subframe segmentation and frequency domain filtering on the high-frequency raw inertial data stream collected by the positioning object based on the preset feature frequency parameters, extract the vibration intensity identifier and generate the vibration-resistant pre-integrated inertial dataset. S3. Input the channel impulse response feature set into the non-line-of-sight classification model to identify the current multipath interference environment and generate a non-line-of-sight environment status identifier. When the non-line-of-sight environment status identifier is generated, call the corresponding error compensation algorithm to calculate and output the final radio frequency observation vector and the penalized observation noise covariance matrix. S4. Convert the vibration intensity label into an expansion coefficient to dynamically adjust the dynamic process noise covariance matrix; combine the adjusted dynamic process noise covariance matrix with iterative calculation to generate the prediction error state covariance matrix; input the prediction error state covariance matrix, the final radio frequency observation vector, the penalized observation noise covariance matrix, and the anti-vibration pre-integration inertial dataset into the error state Kalman filter to perform measurement update and combination calculation, and output the final positioning coordinates; S5. Map the final positioning coordinates to the preset physical virtual model, and construct a dynamic restricted area based on the physical radiation properties of the physical virtual model. When the positioning coordinates intrude into the dynamic restricted area, output an over-limit warning signal and feedback control flow for the positioning object based on the spatial intersection operation result.

[0018] In the preferred embodiment, step S1 specifically includes the following steps: S11. Physical layout topology and shielding of sensors resistant to strong electromagnetic interference. First, based on the spatial radiation parameters of the hydro-generator unit, which include the sector distribution range of stator and rotor magnetic field radiation intensity. The radiation distribution boundary of the strong alternating magnetic field is determined. For the sensing device carried by the target to be located, based on the low-frequency strong interference characteristics of the radiation distribution boundary, a high-permeability permalloy is installed inside its shell as an electromagnetic isolation layer, encapsulating it to form an anti-magnetic interference positioning tag. ; At the same time, according to the physical structure diagram of the factory building The candidate installation locations of UWB base station anchor points are compared with spatial radiation parameters. Perform mapping comparisons to filter out fields with strengths below a preset safety threshold. The designated area is the safe deployment zone, avoiding the strong magnetic radiation main lobe sector extending tangentially from the stator coils. Within this safe deployment zone, UWB base station anchor points are fixed, and their installation positions are recorded, forming the initial physical node state of the UWB base station anchor points. ; Compared to existing technologies, it concretizes the resistance to magnetic field interference from both physical structure and spatial topology dimensions into the addition of shielding materials and safe site selection based on radiation parameters, thereby reducing the physical pollution of the underlying data of the positioning system caused by the strong rotating magnetic field alternation unique to hydropower plants. S12. Construct a smooth UWB anchor point junction coordinate system for environments with significant height differences across floors. Using factory building elevation data The areas of stairwells or through shafts with abrupt changes in Z-axis elevation are extracted and marked as spatial transition zones. Taking the known absolute reference point on the ground floor of the factory building as the origin, combined with... Establish a unified global spatial coordinate system ; Subsequently, the initial physical node states obtained in step S11 are traversed. Map the coordinate system of all UWB base station anchor points to the global spatial coordinate system. In the middle; for those falling into the spatial transition zone Anchor points within, based on Perform overlap coverage checks at vertical and horizontal elevation angles, aggregate all base station anchor points that have completed coordinate calibration and handover area attribute labeling, and generate a set of UWB base station anchor points with absolute spatial reference information. ; Traditional positioning methods often cause trajectory discontinuity problems due to Z-axis coordinate jumps caused by floor switching in complex structures with large elevation differences. This method introduces elevation data to extract transition zones and maintains the continuity of multi-level spaces from the geometric model level. S13. Initialize the time coordination of the multi-source sensors and the system baseline state. After detecting the location tag After being activated upon power-on, a precise time protocol mechanism is used to... The continuous sampling clock of the internal IMU is hardware time-aligned with the ranging clock of the UWB to generate an aligned timestamp; Within the static state window after system startup, via The internal IMU acquires the first segment of silent data, calculates and extracts the accelerometer zero bias and gyroscope zero bias; simultaneously, it utilizes... Receive from UWB base station anchor point set The ranging pulse, combined with Each anchor point in the global spatial coordinate system Given the coordinates below, solve for The initial three-dimensional coordinates and initial attitude; Finally, the calculated initial 3D coordinates, initial attitude, and the extracted IMU zero-bias data are aggregated to construct and initialize the system's baseline state variable set. This serves as the starting point for subsequent mathematical iterations of high-frequency trajectory estimation and ESKF filter updates based on IMU. This ensures time alignment and static reference correction of heterogeneous sensors in complex electromagnetic environments before data fusion, establishing initial state values ​​for subsequent high-frequency dynamic compensation.

[0019] In the preferred embodiment, step S2 specifically includes the following steps: S21. Extract the Channel Impulse Response (CIR) feature set based on the underlying physical signal of UWB. By positioning tags Acquired raw UWB radio frequency signals During the digital baseband processing stage, the raw UWB radio frequency signal Perform waveform analysis and calculation to extract the underlying channel impulse response (CIR) waveform data. ; Subsequently, the channel impulse response waveform data was traversed. By using threshold detection and peak search algorithms, specific physical waveform features are separated, including: the peak attenuation slope feature characterizing the multipath attenuation rate. The first-diameter arrival power characteristic characterizes the energy of the direct path. and the total received power characteristic that characterizes the overall energy of the signal. ; Finally, the extracted peak attenuation slope features Power characteristics of initial diameter and total received power characteristics Tensor splicing and normalization are performed to construct a UWB low-level feature set containing deep multipath micro-motion information. As the judgment input for the subsequent NLOS error dynamic identification model, it distinguishes between dynamic metal reflection and static wall reflection; Compared to existing technologies, this method delves into the underlying CIR waveform, providing high-resolution data input for the subsequent precise removal of dynamic reflection multipath effects from large metal units in hydropower plants; S22. Perform IMU subframe segmentation and pre-filtering to eliminate specific turbine frequency interference. Receive high-frequency raw IMU data stream Subsequently, considering the severe resonance present in the foundation of the hydropower plant, the rated rotational frequency parameters of the turbine generator unit were used. Calculate its corresponding fundamental harmonic period. ; The fundamental harmonic period As a time slice threshold, it is used for continuous high-frequency raw IMU data streams. Subframe segmentation is performed on the time axis, dividing it into several discrete IMU subframe sequences. Next, for each segmented subframe, the initial zero-bias data of the IMU is first subtracted. Zero-bias compensation is performed; then, a preset notch filter is used to process the compensated IMU subframe sequence. Frequency domain filtering is performed when parameters related to the rated rotational frequency of the hydro-generator unit are detected. When the high-frequency noise energy peaks that match its harmonics are detected, the vibration noise in that frequency band is removed. After filtering and denoising, the root mean square value of the residual noise energy in the current subframe is calculated and quantized as a vibration intensity indicator. ; Finally, pre-integration accumulation is performed on the denoised subframe data to generate an anti-vibration pre-integration IMU dataset. This is used for high-frequency inertial calculations of subsequent state update equations; This method addresses the industry's technical bias that IMU data is unreliable when UWB fails by creatively introducing a subframe pre-filtering mechanism based on a specific machine frequency. This effectively prevents the rapid divergence of high-frequency errors generated by traditional IMU integration in a short period of time in the resonant environment of a hydropower plant, thus strengthening the calculation benchmark during the interruption of flow.

[0020] In the preferred embodiment, step S3 specifically includes the following steps: S31. Dynamic NLOS Scene Classification and Recognition Based on Multi-Path Physical Features UWB low-level feature set The pre-built non-line-of-sight multilayer perceptron (MLP) classification model is loaded for feature forward computation. Within the internal logic of the classification model, the following steps are called: A deep analysis of the included physical characteristics is conducted: First, the peak attenuation slope characteristics are calculated. The time-dimensional attenuation gradient is used to determine the stiffness properties of multipath reflecting media; simultaneously, the power characteristics arriving at the first path are calculated. With total received power characteristics The dynamic ratio between them is used to extract the micro-Doppler period jitter caused by the movement of the reflecting surface; the model integrates the above features to make a judgment on the current radio frequency reflection environment: if the evaluation... If a signal exhibits a fixed slope and the power ratio shows no periodic fluctuations, it is classified as "static concrete NLOS"; if high-frequency power ratio fluctuations caused by the Doppler effect are detected, it is classified as "dynamic metal NLOS"; if the signal shows no attenuation and the power is concentrated, it is classified as "line-of-sight (LOS)". Based on the final classification results, the classification model generates the corresponding NLOS environmental status label. , which serves as the control condition for selecting a specific error compensation strategy; Compared to existing technologies, this method distinguishes between different pollution sources caused by "static concrete load-bearing walls" and "dynamic high-speed rotating units" in hydropower plants by analyzing the microscopic characteristics of the input channel impulse response, rather than by simple threshold judgment. S32. Perform UWB ranging error compensation and observation weight calculation based on dynamic classification status. By reading the original UWB ranging vector With NLOS environment status identifier ,according to The three different classification values ​​will execute the corresponding multi-path branching process: when When it is "LOS" (Location of View), maintain With the numerical value unchanged, constant variance is allocated. when When the condition is "Static Concrete NLOS", the preset Support Vector Regression (SVR) operator is invoked to deduct. Due to the fixed time delay caused by penetrating building materials, a smoothed and corrected distance measurement value is obtained, and its variance is appropriately increased. when When the "Dynamic Metal NLOS" occurs (i.e., a severe unit multipath effect occurs), the operation is abandoned. Direct numerical correction is performed, and the limiting observation noise penalty weight is calculated based on the degree of feature distortion. (The extremely large observation variance multiplier) suppresses the outlier data; The final ranging values ​​calculated from the above branch processing are encapsulated to generate the final UWB observation vector. At the same time, the corresponding variance and penalty weights are... Encapsulation generates the observation noise covariance matrix ; To avoid local anomalies in ranging values ​​from causing deviations in the global system positioning trajectory and to ensure system robustness; S33. Construction of time-series alignment and fusion queue for cross-modal sensor data. Because the processing of S31 and S32 causes nonlinear computational delays in UWB data, the system extracts... The hardware receives the timestamp and iterates through the history cache within the timestamp alignment buffer. Sequence; using a sliding window interpolation algorithm, in Match and extract from sequence Timestamp-aligned IMU pre-integrated data slices; After matching, those with the same time base will be... , And the extracted Data slices are encapsulated into structures to generate cross-modal synchronous fusion observation queues. The prediction and update process of the driving error state Kalman filter (ESKF) compensates for the time misalignment of traditional asynchronous sensor data after compensation by complex deep network (MLP), ensuring that the underlying high-frequency inertial estimation and external position observation are time-aligned before entering the Kalman filter.

[0021] In the preferred embodiment, step S4 specifically includes the following steps: S41. IMU Prediction and Estimation Based on Engine Frequency Vibration State Feedback and Adaptive Adjustment of Process Covariance Obtain the set of baseline state variables The previous time step final correction of the state variable and the current vibration-resistant pre-integral IMU dataset ;use right The nominal kinematic equations are derived using Newton-Euler integrals to calculate and generate the predicted nominal state at the current moment. Simultaneously, the system reads the vibration intensity indicator. Construct the dynamic process noise covariance matrix ; Specifically, when When the numerical indication system is in the high-frequency vibration state unique to hydro-generator units, an exponential mathematical attenuation function is used to... Mapped to the expansion coefficient, for The corresponding weights are dynamically amplified to reduce the filter's confidence in the IMU integration path under the current high-frequency vibration; subsequently, this amplified and adjusted weights are combined with... The error state covariance propagation equation is iteratively derived, and the prediction error state covariance matrix is ​​calculated and generated. This provides a priori reference benchmark for updating UWB observations; By introducing In the underlying covariance propagation stage, the weight of the IMU affected by machine frequency contamination is dynamically reduced to isolate the risk of calculation divergence caused by plant resonance from the mathematical model. S42. UWB Measurement Update and Kalman Gain Calculation Based on Dynamic Metal NLOS Penalty Weights Analysis of cross-modal synchronous fusion observation queue Extract the final UWB observation vector Covariance matrix of observation noise Based on predicted nominal state The predicted observation vector is generated by mapping the measurement equation, and then compared with the extracted final UWB observation vector. Perform the difference calculation to obtain the observed residual vector; Next, the prediction error state covariance matrix is ​​used. With the extracted observation noise covariance matrix Perform matrix inversion to calculate and generate the current Kalman gain. It should be specifically pointed out that, due to In the preceding steps, a significant penalty weight has been injected into the "Dynamic Metal NLOS" generated by the hydroelectric metal unit. During this inverse process, the generated Kalman gain is calculated. It will be adaptively compressed, thereby shielding the observation input that is severely contaminated by multipath at the algorithm level; Finally, the generated Kalman gain will be calculated. Multiply by the observed residual vector to generate the optimal error state vector. ; Purification through pre-order NLOS features and The dynamic compression operator enables cross-step linkage, allowing the filter to intelligently and smoothly reject contaminated data when faced with dynamic occlusion of the metal curved surface of a large hydro generator. S43, Global State Combination, Zero-Bias Feedback Correction, and Final High-Precision Coordinate Output Receive Predicted Nominal Status and error state vector Perform a generalized addition operation, and... The position error, velocity error, attitude error, and zero bias error, among other sub-state variables, are injected and compensated for. The corresponding nominal components are combined to generate the final corrected state variable. ; Simultaneously, the prediction error state covariance matrix is ​​utilized. Kalman gain and the observation noise covariance matrix Substituting into the standard covariance update equation, the updated posterior error state covariance matrix is ​​calculated and generated. ; After completing the above combination, the final modified state variable is then... Extract the three-dimensional spatial coordinate matrix and solve it into the final positioning coordinates. The high-precision positioning results of this invention are provided for use by upper-layer business systems; finally, the state variables are corrected. Will be used as the next filtering cycle Feedback is sent to step S41 to form a logical closed loop for continuous positioning; ensuring that the system outputs a stable and continuous high-frequency global trajectory in complex hydropower plant environments.

[0022] In the preferred embodiment, step S5 specifically includes the following steps: S51. Digital twin mapping of hydropower plant based on final positioning coordinates At the system business logic layer, the final location coordinates are received. The data is then imported into a 3D digital twin rendering engine; within the engine, the factory building elevation data is analyzed. And a 3D solid model of the hydropower plant A virtual reference space mapping mesh is constructed; subsequently, an affine transformation of the geometric matrix is ​​performed to convert the mesh into a mesh containing precise three-dimensional coordinate values. Projected onto the mapped grid, making it consistent with The internal spatial topology generates a mount, transforming abstract numerical coordinates into highlighted solid particles that can be tracked in real time in the virtual factory model, rendering and generating a 3D trajectory mapping map of the factory with a spatial location visualization representation. This addresses the technical shortcomings of traditional two-dimensional planar diagrams, which cannot intuitively represent the actual spatial location of personnel in the multi-layered and complex structure of hydropower plants. S52. Dynamic collision avoidance and over-limit warning based on specific strong magnetic field and rotating unit environment. During the background monitoring process, the 3D trajectory mapping map of the factory building is continuously extracted. Real-time updated sequence of entity particle coordinates; synchronous analysis of spatial radiation parameters of the hydro-generator unit. With the preset security electronic fence boundary Based on the structural characteristics of the factory building, a dynamic three-dimensional restricted area model with strong magnetic hazard attributes (such as being close to the strong magnetic field area of ​​the generator stator) and fall hazard attributes (such as being close to the opening of a deep well) is constructed. Next, a spatial Boolean intersection operation is performed between the entity particle coordinate sequence and the dynamic three-dimensional restricted area model. When the operation detects that the particle coordinates have intruded into the dynamic three-dimensional restricted area, a corresponding over-limit warning signal is generated based on the spatial depth of the intrusion and the level of danger attribute through logical branch discrimination. ; Subsequently, the over-limit warning signal is packaged into a protocol to generate a command that can be recognized by the underlying hardware and sent to the control flow. Send the data to the configured location tag via wireless network. This triggers a vibration or buzzer alarm on the wearer; By linking the positioning results with the unique strong magnetic field hazard source and the large drop hazard source of the building in the hydropower plant, a spatial safety active defense logic based on high-precision coordinates is constructed, giving the system real industrial safety production management value. S53. Instantiation and configuration of physical execution components and computer-readable storage media Control the logic flow of the algorithm The code is compiled into a set of binary instructions that can be recognized by the underlying machine hardware, and then burned into the input non-volatile memory element. When the system is powered on and running, the physical processor element drives the input to address and read the binary instruction set from the non-volatile memory element in real time, performing instruction parsing and scheduling. Following a timing sequence, it automatically controls the acquisition of multi-source data, NLOS-specific preprocessing, error compensation, and ESKF filtering, ultimately instantiating a complete fusion high-precision positioning physical execution carrier. .

[0023] In a preferred embodiment, based on the same inventive concept as the aforementioned method, this application also provides a UWB and IMU fusion high-precision positioning system for complex hydropower plant environments, specifically including: The baseline initialization module is used to configure sensor antimagnetic shielding and installation topology constraints, and to construct the global spatial coordinate system of the plant. Based on a precise time protocol, hardware-level time synchronization between UWB and IMU is achieved, and finally, a set of reference state variables containing the zero bias and initial position of multiple source sensors is calculated. ; The anti-interference preprocessing module is used to parse the raw UWB radio frequency signal and extract the channel impulse response (CIR) feature set. Simultaneously, subframe frequency domain filtering is performed on high-frequency IMU data based on the rotation frequency of the hydro-generator unit to quantify and extract vibration intensity indicators. And generate a vibration-resistant pre-integrated IMU dataset. ; Error compensation and alignment module, used to integrate feature sets A multilayer perceptron (MLP) is loaded to identify non-line-of-sight (NLOS) classification states, and the penalized observation noise covariance matrix is ​​calculated accordingly. The observation vectors are corrected, and the data from each modality are packaged into a cross-modal synchronous fusion observation queue using a timestamp alignment algorithm. ; Tightly coupled solution module, used to identify vibration intensity This is converted into an expansion coefficient to dynamically adjust the noise covariance of the prediction process, and combined with the fusion observation queue. The measurement update is performed using Error State Kalman Filtering (ESKF), and the global state combination is finally completed to output high-precision final positioning coordinates. ; The twin mapping and early warning module is used to map the final positioning coordinates. Affine mapping to the 3D solid model of the factory A visual trajectory is generated, and a dynamic three-dimensional restricted area is constructed by combining the unit's strong magnetic radiation parameters and elevation boundaries. The collision avoidance warning signal and the underlying feedback control flow are triggered through spatial Boolean operations.

[0024] Example 2 The comparative example, "A Real-Time UWB Positioning Method and System for Obstacle-Occluded Scenarios" (CN119110244A), constructs a UWB and IMU data fusion architecture based on an Error State Kalman Filter (ESKF) and a Levenberg-Marquardt nonlinear optimization algorithm. Its applicable scenarios and design intentions are mainly aimed at the working conditions in conventional industrial or warehouse environments where UWB signals suffer from non-line-of-sight (NLOS) errors due to static obstacle occlusion. Its key mechanism is to fuse UWB observations and IMU calculations under normal conditions. When the system determines that UWB has NLOS occlusion, the algorithm logic tends to reduce or isolate the measurement weight of UWB and instead rely entirely on the inertial calculations of the IMU to maintain the continuity of the positioning track.

[0025] The underlying logic of the comparative analysis system is based on a common industry assumption: "When UWB fails due to obstruction, IMU data is a reliable relative benchmark." However, in the operating conditions of a hydroelectric power plant, the operation of tens of megawatts of hydroelectric generators generates high-frequency mechanical vibrations and strong alternating magnetic fields. When a target approaches the generator, causing severe UWB multipath distortion, the comparative analysis logic will forcibly switch to IMU dominance. But at this time, the MEMS structure of the IMU is under high-frequency mechanical vibration interference, and its raw accelerometer and gyroscope data already contain a great deal of high-frequency noise. Direct integration and extrapolation will cause the trajectory to diverge exponentially in a short period of time. In addition, the comparative analysis system cannot distinguish between static wall obstruction and dynamic high-speed rotating metal surface obstruction. The use of an absolute isolation strategy is prone to the false negativeing ​​of effective observation data or the omission of contaminated data.

[0026] This embodiment addresses the unidirectional logic of an absolutely trusted IMU by introducing the rated rotation frequency of the turbine unit. By performing subframe frequency domain filtering, the system performs machine frequency denoising on the IMU input during UWB failure, and uses the MLP classification model to analyze the microscopic physical characteristics of CIR, giving the system the ability to identify the material and motion state of multipath interference sources.

[0027] At the computational level, this case no longer uses fixed system process noise; when unit resonance is detected, As a penalty factor, it directly affects the state covariance matrix of the prediction error. The propagation equation actively weakens the system's confidence in the current stimulated IMU prediction state; in the observation link, It is only activated when the classification model determines it to be "Dynamic Metal NLOS", and the expansion observation noise covariance matrix is ​​used. Dynamic variables collectively contribute to the Kalman gain. In the inversion operation, the filter is able to suppress contamination weights under the harsh conditions of UWB multipath and IMU oscillation, rather than simply switching between two unreliable sources.

[0028] To objectively and intuitively quantify the performance difference between the advanced embodiment of this case and the comparative example (CN119110244A) under specific complex operating conditions, this study is specifically based on a closed plant space containing multiple layers of reinforced concrete load-bearing walls and a 50MW-class hydro-turbine generator unit, with the unit's rated rotational frequency... At 2.5Hz / 150rpm, the following data table was obtained to test the position prediction bias, system stability, and anomaly recovery performance caused by the difference in algorithm architecture when facing progressively escalating physical environmental disturbances:

[0029] Under ideal line-of-sight conditions, both exhibit consistent positioning accuracy. However, when encountering situations specific to hydropower plants, the comparative example, when facing dynamic metal multipath, suffers from a lack of underlying waveform classification mechanisms, leading to erroneous high-frequency jumps being treated as valid observations, increasing the error to 0.85m. In contrast, this embodiment relies on dynamic NLOS scene classification based on multipath physical characteristics, effectively shielding distorted signals through penalty weights, stabilizing the error at 0.22m. Under full-load dual interference, the comparative model excessively shifts weights to the IMU when UWB is limited. However, in a real strong resonance environment, this causes its RMSE to exceed 2.5m and triggers filter divergence. The bidirectional weight reduction protection mechanism in this embodiment neither blindly trusts UWB nor blindly trusts the IMU, but extracts the signal through pre-filtering of the generator frequency. This is then incorporated into the propagation calculation of the process noise covariance matrix; the data show that this mechanism of tolerating local uncertainties by expanding the covariance successfully constrains the RMSE within a safe boundary of 0.28m. The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A high-precision positioning method combining UWB and IMU for complex hydropower plant environments, characterized by: Includes the following steps: S1. Based on the preset electromagnetic isolation mechanism and spatial transition zone, determine the initial physical node topology of the positioning object and base station anchor point, establish a global spatial coordinate system and initialize the system's reference state variable set. S2. Analyze the raw radio frequency signal collected by the positioning object to extract the channel impulse response feature set; at the same time, perform subframe segmentation and frequency domain filtering on the high-frequency raw inertial data stream collected by the positioning object based on the preset feature frequency parameters, extract the vibration intensity identifier and generate the vibration-resistant pre-integrated inertial dataset. S3. Input the channel impulse response feature set into the non-line-of-sight classification model to identify the current multipath interference environment and generate a non-line-of-sight environment status identifier. When the non-line-of-sight environment status identifier is generated, call the corresponding error compensation algorithm to calculate and output the final radio frequency observation vector and the penalized observation noise covariance matrix. S4. Convert the vibration intensity label into an expansion coefficient to dynamically adjust the dynamic process noise covariance matrix; combine the adjusted dynamic process noise covariance matrix with iterative calculation to generate the prediction error state covariance matrix; input the prediction error state covariance matrix, the final radio frequency observation vector, the penalized observation noise covariance matrix, and the anti-vibration pre-integration inertial dataset into the error state Kalman filter to perform measurement update and combination calculation, and output the final positioning coordinates; S5. Map the final positioning coordinates to the preset physical virtual model, and construct a dynamic restricted area based on the physical radiation properties of the physical virtual model. When the positioning coordinates intrude into the dynamic restricted area, output an over-limit warning signal and feedback control flow for the positioning object based on the spatial intersection operation result.

2. The high-precision positioning method using UWB and IMU fusion for complex hydropower plant environments as described in claim 1, characterized in that, In step S1, based on the preset electromagnetic isolation mechanism and spatial transition zone, the initial physical node topology of the positioning object and base station anchor point is determined, and a global spatial coordinate system is established, including: The radiation distribution boundary of the alternating magnetic field is determined based on the spatial radiation parameters of the hydro-generator unit, and a high permeability shielding material is set inside the casing of the sensing device carried by the positioning object as an electromagnetic isolation layer. Based on the physical structure diagram of the factory, areas with field strength below the preset safety threshold are selected as safe deployment zones. Strong magnetic radiation sectors are avoided, and base station anchor points are fixed in the safe deployment zones to form the initial physical node state. Using the factory building elevation data, the through area with abrupt elevation change on the Z-axis is extracted and marked as a spatial transition zone. When the base station anchor point falls into the spatial transition zone, the overlap coverage check of the vertical pitch angle is performed based on the elevation data to generate a set of base station anchor points with absolute spatial reference information, and a global spatial coordinate system is established accordingly.

3. The high-precision positioning method using UWB and IMU fusion for complex hydropower plant environments as described in claim 1, characterized in that, Step S1, which establishes a global spatial coordinate system and initializes the system's reference state variable set, also includes a multi-source time coordination and state initialization step, specifically including: After detecting that the positioning object is in a power-on active state, the precise time protocol is invoked to perform hardware-level time alignment between the inertial data sampling clock and the radio frequency ranging clock of the positioning object, and generate an alignment timestamp; Within the preset static state window, the control positioning object collects the first segment of silent inertial data, calculates and extracts the initial zero bias data of the inertial sensor; The initial spatial coordinates and initial attitude of the positioning object are calculated by using the ranging pulse received from the base station anchor point and combining the global spatial coordinate system. The initial zero-bias data, initial spatial coordinates, and initial attitude are aggregated and encapsulated to generate a set of baseline state variables, which serve as the starting point for dynamic solution iteration.

4. The high-precision positioning method using UWB and IMU fusion for complex hydropower plant environments as described in claim 1, characterized in that, In step S2: In the digital baseband processing stage, the original radio frequency signal is analyzed to separate the peak attenuation slope feature that characterizes the multipath attenuation rate, the first path arrival power feature that characterizes the direct path energy, and the total received power feature that characterizes the overall signal energy. The features are then spliced ​​together to construct a set of channel impulse response features. The rated rotational frequency of the hydro-generator unit is used as the characteristic frequency parameter and the corresponding fundamental harmonic period is calculated. The fundamental harmonic period is used as the time slice threshold to divide the high-frequency raw inertial data stream into discrete inertial subframe sequences. The inertial subframe sequence is filtered using a preset frequency domain filtering operator. When a high-frequency noise energy spike that matches the characteristic frequency parameter and its harmonics is detected, the vibration noise in that frequency band is removed. The energy characteristics of residual noise in the filtered subframe are calculated and quantified into vibration intensity indicators. At the same time, the denoised subframe data is pre-integrated and accumulated to generate a vibration-resistant pre-integrated inertial dataset.

5. The high-precision positioning method using UWB and IMU fusion for complex hydropower plant environments as described in claim 1, characterized in that, In step S3: The channel impulse response feature set is loaded into a preset neural network classification model to calculate the time dimension attenuation gradient of the peak attenuation slope feature, and to calculate the dynamic ratio between the first path arrival power feature and the total received power feature. Based on the periodic jitter characteristics of the ratio between the constant state and dynamic value of the attenuation gradient, the multipath interference environment is determined to be either static occlusion or dynamic metallic occlusion, and the corresponding non-line-of-sight environment status label is generated. When the non-line-of-sight environment status indicator is characterized as static occlusion, a preset nonlinear regression compensation operator is invoked to subtract the penetration time delay in the ranging data to generate a smoothed and corrected ranging value, and the variance is increased simultaneously. When the non-line-of-sight environment status indicator is characterized as dynamic metallic occlusion, a limit observation noise penalty weight is calculated based on the degree of signal distortion to suppress the ranging data, and the final ranging value is encapsulated into a final radio frequency observation vector. The variance and the limit observation noise penalty weight are encapsulated to generate a penalized observation noise covariance matrix.

6. A high-precision positioning method for UWB and IMU fusion in complex hydropower plant environments according to claim 1, characterized in that, In step S3, after outputting the final radio frequency observation vector and the penalized observation noise covariance matrix, a time-series alignment step for cross-modal data is also included, specifically: Extract the hardware receiving timestamp corresponding to the final radio frequency observation vector, and traverse the buffered anti-vibration pre-integrated inertial dataset within the preset timestamp alignment buffer; Using a sliding window interpolation algorithm, pre-integrated data slices aligned with hardware receiving timestamps are matched and extracted from the vibration-resistant pre-integrated inertial dataset. The final radio frequency observation vectors with the same time base, the penalized observation noise covariance matrix, and the extracted pre-integrated data slices are encapsulated to generate a cross-modal synchronous fusion observation queue.

7. The high-precision positioning method using UWB and IMU fusion for complex hydropower plant environments according to claim 1, characterized in that, In step S4, the vibration intensity label is converted into an expansion coefficient to dynamically adjust the dynamic process noise covariance matrix. The prediction error state covariance matrix is ​​generated by iterative calculation using the adjusted dynamic process noise covariance matrix, including: Obtain the last corrected state variable from the reference state variable set at the previous moment, and use the anti-vibration pre-integrated inertial dataset to perform dynamic integral derivation of the nominal kinematic equation in the last corrected state variable at the previous moment, and calculate and generate the predicted nominal state at the current moment. The vibration intensity label is input into a preset covariance dynamic weighting function for feature mapping to obtain the corresponding expansion coefficient; The corresponding weights in the dynamic process noise covariance matrix are dynamically amplified using an expansion coefficient; combined with the amplified and adjusted dynamic process noise covariance matrix, the error state covariance propagation equation is iteratively derived and calculated to generate the prediction error state covariance matrix.

8. The high-precision positioning method using UWB and IMU fusion for complex hydropower plant environments according to claim 1, characterized in that, In step S4, the predicted error state covariance matrix, the final radio frequency observation vector, the penalized observation noise covariance matrix, and the vibration-resistant pre-integrated inertial dataset are input into the error state Kalman filter to perform measurement update and combination calculations, outputting the final positioning coordinates, including: Based on the predicted nominal state, a predicted observation vector is generated by mapping the measurement equation, and the difference between the predicted observation vector and the final radio frequency observation vector is calculated to obtain the observation residual vector. By performing matrix inversion operations on the state covariance matrix of the prediction error and the covariance matrix of the penalized observation noise, and combining the adaptive compression effect of the penalty weight, the Kalman gain after bottom-level shielding compression is calculated and generated. The error state vector is generated by multiplying the Kalman gain by the observation residual vector. The error components in the error state vector are compensated to the corresponding nominal components of the predicted nominal state and combined to generate the final corrected state variable. The variable is then substituted into the covariance update equation to calculate the posterior error state covariance matrix. The spatial three-dimensional coordinate matrix is ​​extracted from the final corrected state variable and the final positioning coordinates are obtained by solving the equation.

9. The high-precision positioning method for UWB and IMU fusion for complex hydropower plant environments according to claim 1, characterized in that, In step S5: The final positioning coordinates are imported into the preset 3D solid rendering engine, and the factory building elevation data and solid virtual model are analyzed to construct a reference space mapping mesh. By performing an affine transformation of the geometric matrix, the final positioning coordinates are projected and mounted onto the mapping mesh, and a 3D trajectory mapping map is generated by rendering. Simultaneously analyze the physical radiation properties and the preset safe electronic boundary to construct a dynamic restricted area with magnetic field hazard properties and elevation hazard properties; The real-time updated final positioning coordinate sequence is subjected to spatial Boolean intersection operation with the dynamic restricted area. When the operation result determines that there is coordinate intrusion, an over-limit warning signal is generated through logical branches based on the spatial depth of the intrusion and the level of danger attribute. The over-limit warning signal is packaged into a protocol and a feedback control flow is generated and sent to the located object to trigger a physical alarm.

10. A high-precision positioning system integrating UWB and IMU for complex hydropower plant environments, characterized in that, include: The baseline initialization module is used to determine the initial physical node topology of the positioning object and the base station anchor point according to the preset electromagnetic isolation mechanism and spatial transition zone, establish a global spatial coordinate system and initialize the system's baseline state variable set. The anti-interference preprocessing module is used to parse the raw radio frequency signals collected by the positioning object to extract the channel impulse response feature set; at the same time, based on the preset feature frequency parameters, it performs subframe segmentation and frequency domain filtering on the high-frequency raw inertial data stream collected by the positioning object, extracts vibration intensity identifiers and generates anti-vibration pre-integrated inertial dataset. The error compensation and alignment module is used to input the channel impulse response feature set into the non-line-of-sight classification model, identify the current multipath interference environment and generate a non-line-of-sight environment status identifier; when the non-line-of-sight environment status identifier is generated, the corresponding error compensation algorithm is called to calculate and output the final radio frequency observation vector and the penalized observation noise covariance matrix. The tightly coupled solution module is used to convert vibration intensity indicators into expansion coefficients to dynamically adjust the dynamic process noise covariance matrix; it combines the adjusted dynamic process noise covariance matrix to perform iterative calculations to generate the prediction error state covariance matrix; it inputs the prediction error state covariance matrix, the final RF observation vector, the penalized observation noise covariance matrix, and the anti-vibration pre-integration inertial dataset into the error state Kalman filter to perform measurement update and combination calculations, and outputs the final positioning coordinates; The twin mapping and early warning module is used to map the final positioning coordinates to a preset physical virtual model and construct a dynamic restricted area based on the physical radiation properties of the physical virtual model. When the positioning coordinates intrude into the dynamic restricted area, the module outputs an over-limit warning signal and feedback control flow for the positioning object based on the spatial intersection calculation result.

Images