Adaptive moment estimation optimization positioning methods, systems, devices and storage media

Abstract

This application provides an adaptive moment estimation optimization positioning method, system, device, and storage medium. The method involves deploying Bluetooth beacons on the tunnel wall and recording their coordinates, initializing the coordinates of an inspection vehicle, and having the vehicle receive the strongest RSSI Bluetooth beacon signal in the vicinity. The measured distance to the beacon is calculated using a logarithmic distance path loss model, an optimization objective function is constructed, the gradient of the objective function is calculated, and outlier gradients are clipped. First-order moment momentum updates, second-order moment learning rate adjustments, and moment estimation bias corrections are completed. The vehicle coordinates are iteratively optimized, and gradient clipping and coordinate updates are repeatedly performed until the gradient reaches the target or the maximum number of iterations is reached. By introducing gradient clipping and first-order moment momentum aggregation strategies, the method effectively filters out the drastic RSSI signal jumps caused by multipath effects in subway tunnels, significantly improving positioning stability in complex signal environments. It achieves centimeter-level high-precision positioning without additional hardware investment and boasts advantages such as low cost, flexible deployment, strong anti-interference capabilities, and high robustness.

Description

Technical Field

[0001] This application relates to the field of indoor positioning technology, and in particular to adaptive moment estimation optimization positioning methods, systems, devices and storage media. Background Technology

[0002] In subway inspection scenarios, underground spaces such as tunnels are signal blind spots for Global Navigation Satellite Systems (GNSS), causing significant cumulative errors to accumulate over time when inspection vehicles rely solely on inertial navigation. RSSI positioning based on Bluetooth beacons is often used as a low-cost supplementary positioning method due to its low cost and flexible deployment. It is used to achieve area-specific awareness or as a calibration point to receive signals and correct the accumulated errors of inertial navigation, thereby achieving continuous fusion positioning.

[0003] In practical applications, RSSI Bluetooth positioning technology has significant limitations. First, in the complex environment of subway tunnels, Bluetooth signal propagation is severely affected by physical obstruction, multipath effects, and non-line-of-sight propagation, resulting in highly fluctuating and unstable RSSI signal measurements, making it difficult to use directly as a high-precision primary positioning solution. Second, when the system uses traditional mathematical optimization algorithms to process these ranging data filled with environmental noise, it faces problems such as slow convergence speed and easy getting trapped in local optima. Furthermore, the fixed learning rate is not adaptable to drastically fluctuating signals, and initial iteration bias and robustness defects also severely restrict the final positioning accuracy. Third, existing algorithm improvements still have shortcomings in momentum optimization and adaptive learning rates. They fail to fully integrate historical gradient information to filter and suppress signal jump noise, the adjustment of the learning rate is difficult to dynamically match the complex signal characteristics of the tunnel environment, and there is a general lack of initial iteration bias correction mechanisms. Summary of the Invention

[0004] The technical problem this application aims to solve is that RSSI Bluetooth positioning technology itself has obvious limitations in practical applications. First, in the complex environment of subway tunnels, Bluetooth signal propagation is affected by factors such as severe physical obstruction, multipath effects, and non-line-of-sight propagation, resulting in large fluctuations and poor stability of the received RSSI signal measurements, making it difficult to directly use as a high-precision primary positioning solution. Second, when the system uses traditional mathematical optimization algorithms to process these ranging data filled with environmental noise, it faces the problems of slow convergence speed and easy getting trapped in local optima; moreover, the fixed learning rate is not adaptable to drastically fluctuating signals, and the initial iteration bias and robustness defects also seriously restrict the final positioning accuracy. Third, existing algorithm improvements still have shortcomings in momentum optimization and adaptive learning rate, failing to fully combine historical gradient information to filter and suppress signal jump noise, the adjustment of the learning rate is difficult to dynamically match the complex signal characteristics of the tunnel environment, and there is a general lack of initial iteration bias correction mechanism.

[0005] To address, or at least partially address, the aforementioned technical problems, this application provides an adaptive moment estimation optimization positioning method, system, device, and storage medium.

[0006] In a first aspect, the present invention discloses an adaptive moment estimation optimization localization method, which includes, The initial parameters are obtained. The vehicle acquires the signal sent by the Bluetooth beacon, selects a preset number of core beacons based on the signal strength, constructs a logarithmic distance path loss model, and calculates the distance between the vehicle and the core beacon by inputting the signal strength of the core beacon to obtain the measured distance. The initial parameters include the tunnel physical parameters, the coordinates of the Bluetooth beacon in the tunnel, and the coordinates of the first vehicle. Based on the coordinates of the first vehicle, the coordinates of the core beacon, and the measured distance, an objective function for the vehicle coordinates and the core coordinates is constructed. Adaptive gradient calculation and adaptive moment update are performed on the objective function to obtain the updated vehicle coordinates, the L2 norm of the gradient vector, and the number of iterations. Repeat the adaptive gradient calculation and adaptive moment update process until the preset gradient threshold exceeds the L2 norm of the gradient vector or the number of iterations exceeds the preset maximum number of iterations, and obtain the optimized car coordinates.

[0007] Preferably, the initial parameters are obtained, the vehicle acquires the signal transmitted by the Bluetooth beacon, selects a preset number of core beacons based on the signal strength, constructs a logarithmic distance path loss model, and calculates the distance between the vehicle and the core beacon by inputting the signal strength of the core beacon, thus obtaining the measured distance. Specifically, this includes the following steps: Obtain tunnel parameters and Bluetooth beacon parameters, install Bluetooth beacons at equal intervals in the tunnel, initialize the vehicle coordinates, and obtain the first vehicle coordinates. Tunnel parameters include tunnel length and Bluetooth beacon coordinates. The car receives Bluetooth beacon wireless signals in real time, sorts them based on signal strength, and selects the beacons with the strongest signal strength as the core beacons. A logarithmic distance path loss model is constructed. The signal strength of the core beacon is input, and the distance between the vehicle and the core beacon is calculated to obtain the measured distance.

[0008] Preferably, based on the coordinates of the first vehicle, the coordinates of the core beacon, and the measured distance, an objective function for the vehicle coordinates and the core coordinates is constructed. Adaptive gradient calculation and adaptive moment update processing are then performed on the objective function to obtain the updated vehicle coordinates, the L2 norm of the gradient vector, and the number of iterations. Specifically, this includes the following steps: Based on the coordinates of the first vehicle, the coordinates of the core beacon, and the measured distance, the theoretical geometric distance between the coordinates of the first vehicle and the coordinates of the core beacon is calculated, and the formula for the theoretical geometric distance is obtained. By combining the theoretical geometric distance and the measured distance, the ranging residual is calculated, and an objective function for positioning optimization is constructed based on the ranging residual, thus obtaining the objective function; The objective function performs adaptive gradient calculation on the coordinates of the first car to obtain the gradient parameters of the objective function. The gradient parameters include the adaptive gradient and the L2 norm of the gradient vector. The adaptive moments are updated based on the gradient parameters of the objective function, and the actual coordinates and iteration number of the car are updated to obtain the updated car coordinates and iteration number.

[0009] Preferably, the objective function undergoes adaptive gradient calculation with respect to the coordinates of the first vehicle to obtain the gradient parameters of the objective function. The gradient parameters include the adaptive gradient and the L2 norm of the gradient vector. Specifically, this includes the following steps: Based on the objective function constructed from the coordinates of the first vehicle, the gradient of the objective function with respect to the vehicle coordinates is calculated at the current iteration number, thus obtaining the gradient of the objective function; Calculate the L2 norm of the gradient of the objective function to obtain the L2 norm of the gradient vector; A preset clipping threshold is set. The clipping threshold is compared with the L2 norm of the gradient. If the L2 norm of the gradient vector is greater than the clipping threshold, the gradient is clipped to obtain the clipped gradient.

[0010] Preferably, the adaptive moments are updated based on the gradient parameters of the objective function, and the actual coordinates and iteration number of the vehicle are updated to obtain the updated vehicle coordinates and iteration number. Specifically, this includes the following steps: Based on the clipped gradient, the first and second moments are updated to obtain the first and second moment parameters. The first-order moment parameters and the second-order moment parameters are corrected for deviation to obtain the corrected first-order moment and the corrected second-order moment. Based on the corrected first and second moments, the car coordinates are iteratively corrected once, and the number of iterations is accumulated once to obtain the updated car coordinates and the number of iterations.

[0011] Preferably, based on the clipped gradient, the first and second moments are updated to obtain the first and second moment parameters, specifically including the following steps: Obtain the first moment obtained from the previous iteration, the pruned gradient in this iteration, and the preset first moment decay factor. Calculate the first moment according to the weighted fusion rule to obtain the first moment parameters. Obtain the first moment obtained from the previous iteration, the gradient after pruning in this iteration, and the preset second moment decay factor. Calculate the squared value of the gradient after pruning in this iteration. Calculate the second moment by combining the first moment obtained from the previous iteration, the squared value of the gradient after pruning in this iteration, and the preset second moment decay factor according to the weighted fusion rule to obtain the second moment parameter.

[0012] Preferably, the adaptive gradient calculation and adaptive moment update process are repeated until the preset gradient threshold exceeds the L2 norm of the gradient vector or the number of iterations exceeds the preset maximum number of iterations, to obtain the optimized car coordinates. Specifically, this includes the following steps: The updated car coordinates are iterated repeatedly. During the iteration, the updated car coordinates are input into the objective function to replace the first car coordinates, and the current iteration process is carried out. The objective function repeatedly undergoes adaptive gradient calculation and adaptive moment update processing to iteratively update the car coordinates; When the preset gradient threshold exceeds the L2 norm of the gradient vector or the number of iterations exceeds the preset maximum number of iterations, stop iterating and updating the car coordinates, and output the optimized car coordinates.

[0013] Secondly, the present invention discloses an adaptive moment estimation optimization positioning system, which includes the above-mentioned adaptive moment estimation optimization positioning method.

[0014] Thirdly, the present invention discloses an electronic device, which includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When the processor executes the program stored in memory, it implements the steps of the adaptive moment estimation optimization positioning method described above.

[0015] Fourthly, the present invention discloses a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the adaptive moment estimation optimization positioning method as described above.

[0016] The technical solution provided in this application has the following advantages compared with the prior art: The adaptive moment estimation optimization positioning method, system, device, and storage medium provided in this application are as follows: The method involves first deploying Bluetooth beacons on the tunnel wall and recording their coordinates, and initializing the coordinates of the inspection vehicle. The vehicle receives signals from 4-6 of the strongest RSSI Bluetooth beacons in the vicinity. The measured distance to the beacons is calculated using a logarithmic distance path loss model. An optimization objective function is constructed, the gradient of the objective function is calculated, and abnormal gradients are clipped. First-order moment momentum updates, second-order moment learning rate adjustments, and moment estimation bias corrections are completed. The vehicle coordinates are iteratively optimized. Finally, gradient clipping and coordinate updates are repeatedly performed until the gradient reaches the target or the number of iterations is reached. If the vehicle gets stuck in a local optimum, the coordinates are reinitialized and the iteration is repeated. Finally, the vehicle's positioning coordinates are output. By introducing gradient pruning and first-order moment momentum aggregation strategies, the drastic changes in RSSI signals caused by multipath effects in subway tunnels are effectively filtered out, significantly improving positioning stability in complex signal environments. Centimeter-level high-precision positioning can be achieved without additional hardware investment. It has the advantages of low cost, flexible deployment, strong anti-interference, and high robustness. It can stably converge to the global optimum in complex scenarios such as GNSS blind spots in subway tunnels and large initial positioning errors, perfectly adapting to the continuous positioning needs of subway inspection vehicles. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating an adaptive moment estimation optimization positioning method provided in this application; Figure 2 A flowchart of an adaptive moment estimation optimization positioning method provided in this application; Figure 3 A flowchart illustrating step S1 of the adaptive moment estimation optimization positioning method provided in this application; Figure 4 A flowchart illustrating step S2 of the adaptive moment estimation optimization positioning method provided in this application; Figure 5 A flowchart illustrating step S2 of the adaptive moment estimation optimization positioning method provided in this application; Figure 6 Simulation trajectory diagram of an adaptive moment estimation optimization positioning method provided in this application; Figure 7 The curvature plot of the average error of sampling points in the adaptive moment estimation optimization positioning method provided in this application decreases with the number of iterations. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Firstly, see Figures 1-7 This invention discloses an adaptive moment estimation optimization localization method, which includes, Step S1: Obtain initial parameters. The vehicle obtains the signal sent by the Bluetooth beacon, selects a preset number of core beacons based on the signal strength, constructs a logarithmic distance path loss model, inputs the signal strength of the core beacon to calculate the distance between the vehicle and the core beacon, and obtains the measured distance. The initial parameters include the tunnel physical parameters, the coordinates of the Bluetooth beacon in the tunnel, and the coordinates of the first vehicle. Step S2: Based on the coordinates of the first vehicle, the coordinates of the core beacon, and the measured distance, construct the objective function of the vehicle coordinates and the core coordinates. Perform adaptive gradient calculation and adaptive moment update on the objective function to obtain the updated vehicle coordinates, the L2 norm of the gradient vector, and the number of iterations. Step S3: Repeat the adaptive gradient calculation and adaptive moment update process until the preset gradient threshold exceeds the L2 norm of the gradient vector or the number of iterations exceeds the preset maximum number of iterations, and obtain the optimized car coordinates.

[0022] Specifically, in step S1, initial parameters such as tunnel physical parameters, Bluetooth beacon coordinates within the tunnel, and initial coordinates of the first vehicle are first acquired. The vehicle receives signals transmitted by the Bluetooth beacons, and a preset number of core beacons are selected based on signal strength. A logarithmic distance path loss model adapted to the tunnel environment is constructed. The signal strength of the core beacons is input into the model to calculate the measured distance between the vehicle and each core beacon, completing the acquisition of basic positioning data and the conversion of signal to distance. This provides accurate ranging input for subsequent coordinate optimization. During the process, the beacon selection rules are simple and efficient, and the ranging model fits the wireless transmission characteristics of the tunnel. Low-cost and highly adaptable ranging calculations can be completed without complex hardware, ensuring the reliability of the basic data.

[0023] Specifically, in step S2, an objective function is constructed based on the coordinates of the first vehicle, the coordinates of the core beacon, and the measurement distance, with the sum of squared distance residuals as the core. First, adaptive gradient calculation is performed on the objective function, and abnormal gradients are suppressed by gradient clipping. Then, adaptive moment update processing is carried out for first-order moment, second-order moment update, and deviation correction. Finally, the updated vehicle coordinates, gradient vector L2 norm, and iteration number are output, transforming the positioning problem into a mathematical optimization problem. The first iteration correction of the vehicle coordinates is completed, and the iteration judgment index is generated. Gradient clipping can filter tunnel signal noise, and adaptive moment update takes into account momentum acceleration and learning rate adaptation. The coordinate correction is more accurate, the iteration convergence is faster, and the stability of the optimization process is effectively improved.

[0024] Specifically, in step S3, the iterative optimization operations of adaptive gradient calculation and adaptive moment update are repeatedly executed to continuously correct the vehicle coordinates. It is determined in real time whether the L2 norm of the gradient vector is less than the preset gradient threshold and whether the number of iterations exceeds the preset maximum number of iterations. If either condition is met, the iteration is terminated and the optimized vehicle coordinates are output. The positioning error is continuously reduced through closed-loop iteration to ensure that the vehicle coordinates converge to the global optimum and complete the final high-precision positioning output. The iteration termination condition is clear and controllable, which takes into account both positioning accuracy and computational efficiency. It can effectively avoid the algorithm from getting trapped in local optima, adapt to the complex environment of tunnel GNSS blind zone, and achieve stable centimeter-level positioning.

[0025] Understandably, this solution first deploys Bluetooth beacons on the tunnel wall and records their coordinates, initializing the coordinates of the inspection vehicle. The vehicle receives signals from 4-6 of the strongest RSSI Bluetooth beacons in the vicinity. It calculates the measured distance to the beacons using a logarithmic distance path loss model, constructs an optimization objective function, calculates the gradient of the objective function, and prunes abnormal gradients. It then completes first-order moment momentum updates, second-order moment learning rate adjustments, and moment estimation bias corrections, iteratively optimizing the vehicle coordinates. Finally, it repeatedly performs gradient pruning and coordinate updates until the gradient reaches the target or the iteration count is reached. If it gets stuck in a local optimum, it re-initializes the coordinates and iterates again, ultimately outputting the vehicle's positioning coordinates. By introducing gradient pruning and first-order moment momentum aggregation strategies, it effectively filters out the drastic jumps in RSSI signals caused by multipath effects in subway tunnels, significantly improving positioning stability in complex signal environments. It achieves centimeter-level high-precision positioning without additional hardware investment, combining advantages such as low cost, flexible deployment, strong anti-interference, and high robustness. It can stably converge to the global optimum in complex scenarios such as GNSS blind spots and large initial positioning errors in subway tunnels, perfectly meeting the continuous positioning needs of subway inspection vehicles.

[0026] Furthermore, to address the issue of large initial coordinate estimation errors in the blind zone, a deviation correction mechanism is adopted to compensate for initial iteration deviations, ensuring that the algorithm can still accurately find the optimal descent direction under adverse initial conditions and improving robustness.

[0027] Furthermore, without the need for additional expensive hardware, the algorithm optimization alone compensates for the physical limitations of Bluetooth ranging, providing cost-effective inertial navigation calibration point support for inspection vehicles in GNSS blind zones.

[0028] Step S1 specifically includes the following steps: Step S11: Obtain tunnel parameters and Bluetooth beacon parameters, install Bluetooth beacons at equal intervals in the tunnel, initialize the vehicle coordinates, and obtain the first vehicle coordinates. The tunnel parameters include the tunnel length and the coordinates of the Bluetooth beacons. Step S12: The car receives the wireless signal of the Bluetooth beacon in real time, sorts the wireless signals based on their signal strength, and selects the beacons with the strongest preset number as the core beacons; Step S13: Construct a logarithmic distance path loss model, input the signal strength of the core beacon, calculate the distance between the vehicle and the core beacon, and obtain the measured distance.

[0029] Specifically, the core physical parameters of the tunnel (including tunnel length and tunnel outline dimensions) and Bluetooth beacon parameters are first collected. A wireless signal receiving module and a positioning module are installed inside the vehicle. Bluetooth beacons are evenly installed on both sides of the tunnel wall according to a fixed equal spacing rule. The two-dimensional physical coordinates of each Bluetooth beacon are accurately measured and recorded. Then, the positioning coordinates of the subway inspection vehicle are randomly initialized to obtain the initial value for iterative optimization. The positioning coordinates of the subway inspection vehicle are the coordinates of the first vehicle. After completing the positioning benchmark and initial parameter configuration, the inspection vehicle receives wireless signals from all deployed Bluetooth beacons in the tunnel in real time via its onboard Bluetooth communication module, extracting the RSSI (Signal Strength Indicator) value for each beacon. All RSSI values ​​are sorted from highest to lowest signal strength, and a preset number (4-6) of the strongest beacons are selected and defined as core beacons. Only the identifiers and RSSI data of the core beacons are retained for subsequent calculations. Based on the characteristics of the tunnel wireless transmission environment, a logarithmic distance path loss model is constructed. The RSSI values ​​of the core beacons obtained in step S12 are substituted into the model formula, and the physical distance between the vehicle and each core beacon is calculated through formula transformation to obtain the measured distance. .

[0030] The specific steps are as follows: The vehicle travels along the track and selects 4 to 6 Bluetooth beacons with the strongest received signal strength. The signal information includes beacon identification and signal measurement strength. The distance between the vehicle and the corresponding beacon was calculated using a logarithmic distance path loss model based on the signal strength of different beacons. ,Right now ; In the formula Let be the signal strength value of the i-th Bluetooth beacon received by the car; It is a loss factor related to the wireless transmission environment; It is necessary to calculate the measured distance between the car and the beacon. This is the absolute reference signal strength when the transmitting and receiving ends are 1 meter apart.

[0031] Step S2 specifically includes the following steps: Step S21: Based on the coordinates of the first vehicle, the coordinates of the core beacon, and the measured distance, calculate the theoretical geometric distance between the coordinates of the first vehicle and the coordinates of the core beacon, and obtain the formula for the theoretical geometric distance; Step S22: Combine the theoretical geometric distance with the measured distance to calculate the ranging residual, and construct the objective function for positioning optimization based on the ranging residual to obtain the objective function; Step S23: The objective function performs adaptive gradient calculation on the coordinates of the first car to obtain the gradient parameters of the objective function. The gradient parameters include the adaptive gradient and the L2 norm of the gradient vector. Step S24: Update the adaptive moments based on the gradient parameters of the objective function, update the actual coordinates and iteration number of the car, and obtain the updated car coordinates and iteration number.

[0032] Specifically, based on the coordinates of the first vehicle and the coordinates of the core beacons, and according to the Euclidean geometry principle of two-dimensional plane, the theoretical geometric distance between the initial estimated position of the vehicle and each core beacon is calculated, and the corresponding theoretical geometric distance calculation formula is obtained. The difference between the theoretical geometric distance and the actual measured distance obtained in step S1 is calculated to obtain the ranging residual. Then, the sum of the squares of the ranging residuals of all core beacons is used as the core to construct the positioning optimization objective function and determine the function formula. The partial derivative of the constructed objective function with respect to the coordinates of the first vehicle is calculated to obtain the original gradient. Then, the L2 norm of the gradient vector is calculated. Gradients exceeding the threshold are clipped to obtain the adaptive gradient. Finally, two types of gradient parameters, namely the adaptive gradient and the L2 norm of the gradient vector, are output. Based on the adaptive gradient parameters, the first moment and the second moment are updated and the bias is corrected in sequence. Then, the vehicle coordinates are iteratively corrected in combination with the basic learning rate. At the same time, the number of iterations is accumulated and updated, and finally the updated vehicle coordinates and the latest iteration number are obtained. Step S2 can effectively quantify the ranging error caused by tunnel signal noise, suppress abnormal gradient fluctuations through gradient clipping, ensure the stability of coordinate optimization direction, and combine adaptive moment update to achieve momentum acceleration and learning rate adaptation, which greatly improves coordinate correction efficiency and convergence speed. Even with large initial coordinate deviations, it can still robustly approximate the true position, taking into account optimization accuracy, anti-interference ability and computational efficiency.

[0033] The specific operation is as follows, using the initial estimated coordinates of the first trolley. Based on the known true coordinates of each core beacon, Based on the principles of two-dimensional Euclidean geometry, the theoretical geometric straight-line distance from the vehicle's current estimated position to each core beacon is calculated. This distance is an ideal distance value free from noise interference. The calculation formula is as follows: ; in, Let be the theoretical geometric distance between the vehicle and the i-th core beacon. The coordinates of the first car; Let be the coordinates of the i-th core beacon. Subsequently, let the coordinates of the subway inspection vehicle to be optimized in the current iteration be . The theoretical geometric distance between the coordinates of the subway inspection vehicle and the coordinates of the i-th core beacon in the iterative calculation is: ; Then, the difference between the theoretical geometric distance and the actual measured distance di obtained in step S1 is used to obtain the ranging residual, which can reflect the ranging error caused by signal noise. Then, the objective function of adaptive moment estimation optimization is constructed with the sum of squared ranging residuals of all core beacons as the core. The car localization problem is transformed into a mathematical optimization problem of minimizing the objective function. The formula for calculating the distance residual is as follows: ; The formula for calculating the objective function is as follows: ; in, Let m be the ranging residual of the i-th core beacon, and m be the number of core beacons (4-6). To optimize the objective function using the initial coordinates of the first car, To optimize the objective function for the car coordinates after the current iteration, the objective function is... The partial derivatives of the vehicle's x-coordinate and y-coordinate are calculated separately to obtain the initial position gradient of the objective function. Subsequently, during iterative calculations, the updated vehicle coordinates replace the initial vehicle coordinates for subsequent iterations. Then, the L2 norm of the gradient vector is calculated based on the objective function. If the gradient exceeds a threshold, gradient clipping is performed to obtain an adaptive gradient. The L2 norm of the gradient vector is output as a criterion for judging iterative convergence. (Original gradient vector) Where g is the original gradient, The gradient component of the x-axis. Let be the gradient component of the ordinate, calculated using the following formula: ; ; The gradient vector L2 norm is calculated based on the original gradient, using the following formula: In the formula, It is the L2 norm of the gradient vector, used for gradient determination and convergence detection; The gradient clipping (adaptive gradient) calculation formula is as follows: , ( > (triggered at time), in addition, when ≤ hour, No changes will be made; the current value will be used.

[0034] in, For adaptive gradient, This is the preset gradient clipping threshold.

[0035] Finally, the first moment (momentum) and second moment (learning rate base) are updated sequentially based on the adaptive gradient. After completing the bias correction, the car coordinates are iteratively corrected in combination with the basic learning rate. At the same time, the iteration count is incremented by 1, and finally the updated car coordinates and the latest iteration count are obtained.

[0036] Step S23 specifically includes the following steps: Step S231: Construct an objective function based on the coordinates of the first vehicle, calculate the gradient of the objective function with respect to the vehicle coordinates at the current iteration number, and obtain the gradient of the objective function; Step S232: Calculate the L2 norm of the gradient of the objective function to obtain the L2 norm of the gradient vector; Step S233: Set a clipping threshold, compare the clipping threshold with the L2 norm of the gradient. If the L2 norm of the gradient vector is greater than the clipping threshold, clip the gradient to obtain the clipped gradient.

[0037] Specifically, the gradient of the objective function is calculated to determine the precise descent direction for vehicle coordinate optimization. The gradient norm 2 is calculated to provide a quantitative standard for gradient anomaly detection and subsequent iteration convergence detection. Gradient clipping constrains abnormal gradients exceeding a threshold, eliminating gradient fluctuations caused by RSSI signal noise in the tunnel environment and ensuring the reliability of the coordinate optimization direction. This method can accurately pinpoint the optimal correction direction for vehicle coordinates. The gradient norm 2 is used to quantify and evaluate gradient fluctuations. The gradient clipping mechanism effectively suppresses abnormal gradient oscillations caused by the complex tunnel environment, preventing coordinate iterations from deviating from the true position due to noise interference, and significantly improving the stability and anti-interference capability of the coordinate optimization process.

[0038] Step S24 specifically includes the following steps: Step S241: Based on the clipped gradient, perform first-order moment and second-order moment update processing to obtain first-order moment parameters and second-order moment parameters; Step S242: Perform deviation correction processing on the first-order moment parameters and the second-order moment parameters to obtain the corrected first-order moment and the corrected second-order moment; Step S243: Based on the corrected first moment and the corrected second moment, perform an iterative correction on the car coordinates, and increment the number of iterations to obtain the updated car coordinates and the number of iterations.

[0039] Specifically, a moment update is first performed. This moment update aggregates historical gradient trends to provide momentum acceleration for coordinate optimization, improves the iteration convergence speed, and aligns with the dynamic trend of positioning error optimization. The formula for the first moment (momentum direction) update is as follows: ; In the formula, It is the number of iterations; Let t be the first moment of the t-th order. The first moment of the previous iteration is the nth moment. The manifestation of the momentum vector in the next iteration It is the first-order moment decay factor. For the first Gradient after clipping in the next iteration .

[0040] Then, the second moment is updated. The second moment update is the cumulative squared feature of the statistical gradient, which provides a basis for dynamically scaling the learning rate, allowing the learning rate to adapt to the complex signal environment of the tunnel and avoiding the insufficient adaptability of a fixed learning rate. The formula for the second moment update is as follows: ; In the formula, It is the second moment of the t-th iteration, used to adaptively adjust the learning rate, and is a statistic related to the learning rate and its cardinality. It is the second moment of the previous iteration. It is the second-order moment decay factor. For the first Gradient after clipping in the next iteration .

[0041] Deviation corrections are applied to the first and second moments to ensure that the momentum estimate accurately reflects the true gradient characteristics of the optimized positioning parameters. The deviation correction calculation formula is as follows: ; ; In the formula The corrected first moment, The first-order moment decay factor raised to the power of t. To correct the second-order moment, It is the t-th power of the first-order moment decay factor.

[0042] Based on the corrected first and second moments, the vehicle coordinates are updated and optimized. The update formula is as follows: ; ; It is the first-order moment decay factor. It is the second-order moment attenuation factor. Based on the learning rate, It is a smoothing constant. The updated car coordinates are used; after each iteration, the iteration count is updated, and the update count is incremented. In the formula, This is the latest iteration number. This represents the number of the previous iteration.

[0043] In step S24, momentum acceleration is achieved by aggregating historical gradient trends using the first moment, and the learning rate is adaptively adjusted by statistically analyzing the squared gradient features of the second moment. Bias correction compensates for the moment estimation error in the initial iteration stage. Then, the vehicle coordinates are accurately corrected based on the corrected moment parameters, and the iteration count is updated synchronously, providing a new coordinate benchmark and counting basis for subsequent iterations, continuously pushing the estimated vehicle coordinates closer to the true position. Updating the first and second moments can balance the iteration convergence speed with the adaptability to the complex signal environment of the tunnel. Bias correction effectively improves the robustness of the algorithm in the initial stage. Single coordinate iteration correction is accurate and efficient, and the accumulation of iteration counts is clear and standardized. Overall, the coordinate optimization achieves rapid convergence and accurate correction, greatly enhancing the stability and anti-interference ability of the positioning algorithm.

[0044] Step S241 specifically includes the following steps: Step S2411: Obtain the first moment obtained in the previous iteration, the pruned gradient in this iteration, and the preset first moment decay factor. Calculate the first moment according to the weighted fusion rule to obtain the first moment parameters. Step S2412: Obtain the first moment obtained from the previous iteration, the gradient after clipping in this iteration, and the preset second moment decay factor. Calculate the squared value of the gradient after clipping in this iteration. Calculate the second moment by combining the first moment obtained from the previous iteration, the squared value of the gradient after clipping in this iteration, and the preset second moment decay factor according to the weighted fusion rule to obtain the second moment parameter.

[0045] Specifically, first, obtain the first moment calculated in the previous iteration, the gradient after pruning in the current iteration, and the preset first moment decay factor. According to the weighted fusion calculation rules, multiply the first moment from the previous iteration by the first moment decay factor, multiply the pruned gradient by (1 minus the first moment decay factor), and then add the two product results to complete the update calculation of the first moment and obtain the current first moment parameter. Next, obtain the second moment obtained in the previous iteration, the gradient after pruning in the current iteration, and the preset second moment decay factor. First, square the gradient after pruning in the current iteration to obtain the gradient square value. Then, according to the weighted fusion rules, multiply the second moment from the previous iteration by the second moment decay factor, multiply the gradient square value by (1 minus the second moment decay factor), and finally add the two product results to complete the update calculation of the second moment and obtain the current second moment parameter. Step S241 updates the first and second moments by weighted fusion of historical moment information and current effective gradient information. The first moment is used to aggregate historical gradient trends to form the momentum direction for iterative optimization, while the second moment is used to statistically calculate the cumulative squared features of the gradient, providing a computational basis for the adaptive learning rate for subsequent coordinate correction, and providing core parameter support for bias correction and coordinate iterative updates. Updating the moment parameters using a weighted fusion method preserves the effective trend information of historical gradients while incorporating current stable gradient data. The first moment update imbues the coordinate iteration with momentum characteristics to accelerate convergence, while the second moment update accurately senses gradient fluctuations, effectively adapting to the complex signal environment of tunnels, significantly improving the accuracy and anti-interference capabilities of the moment parameters, and laying the foundation for subsequent accurate correction of the vehicle's coordinates.

[0046] Step S3 specifically includes the following steps: Step S31: Repeatedly iterate the updated car coordinates. During the iteration, input the updated car coordinates into the objective function to replace the first car coordinates and perform the current iteration process. Step S32: Repeatedly perform adaptive gradient calculation and adaptive moment update on the objective function to iteratively update the car coordinates; Step S33: When the preset gradient threshold exceeds the L2 norm of the gradient vector or the number of iterations exceeds the preset maximum number of iterations, stop iterating and updating the car coordinates, and output the optimized car coordinates.

[0047] Specifically, the updated car coordinates obtained from the previous iteration are substituted into the positioning objective function, replacing the initial first car coordinates as the calculation benchmark for this iteration. The entire process of adaptive gradient calculation, gradient vector L2 norm calculation, and adaptive moment update is repeatedly executed to continuously iterate and correct the car coordinates. During the iteration process, the gradient vector L2 norm is compared with the preset gradient threshold, and the current iteration number is compared with the preset maximum iteration number. When the preset gradient threshold is greater than the gradient vector L2 norm or the number of iterations exceeds the maximum number of iterations, the iteration is stopped immediately. If the number of iterations reaches the maximum number of iterations but the current objective function value is still greater than the preset gradient threshold, the algorithm is determined to be trapped in a local optimum. The system automatically reinitializes the car coordinates to be located and restarts the complete iterative calculation process until it converges to the global optimum, and finally outputs the optimized precise car coordinates. Closed-loop iteration can steadily improve positioning accuracy. The dual termination conditions balance positioning accuracy and computational efficiency, avoiding meaningless iterations that waste computing power. The local optimum detection and re-initialization mechanism greatly enhances the robustness of the algorithm, effectively coping with harsh scenarios such as complex signal noise in tunnels and large initial coordinate deviations, ensuring that the algorithm always converges to the global optimum, and finally outputs stable and reliable centimeter-level vehicle positioning coordinates, which fully meet the positioning requirements of subway inspection.

[0048] That is, adaptive gradient calculation and adaptive moment update are repeatedly performed, iterating until one of two conditions is met: one is that the number of iterations exceeds the preset maximum number of iterations, and the other is that the latest iteration number is detected. Has the preset maximum number of iterations been exceeded? (When the latest iteration count...) The iteration process stops once the preset maximum number of iterations is exceeded. In this embodiment, the upper limit of the iteration count t is 2000, but the maximum number of iterations is not limited to the number set in this embodiment. Another condition is that the L2 norm of the gradient of the current objective function with respect to the vehicle's coordinates is less than a set gradient threshold. ≤ hour, The current value remains unchanged, and the iterative processing stops. In this embodiment, the preset gradient threshold is 0.001, but the gradient threshold is not limited to the value set in this embodiment. Furthermore, if the number of iterations reaches the preset maximum number of iterations, and the current objective function value is still greater than the preset gradient threshold, it is determined that the current optimization has fallen into a local optimum. The system reinitializes the coordinates of the vehicle to be located and restarts the iterative calculation to ensure convergence to the global optimum.

[0049] In one embodiment, such as Figure 6 As shown, the experimental map is two-dimensional and 30 meters long. A smooth S-shaped tunnel with a width of 4 meters and a total central path length of approximately 71.4 meters is constructed within a 30-meter plane. Bluetooth beacons are evenly spaced at 2-meter intervals along both tunnel walls, and 80 path sampling points are set along the tunnel's centerline. In the experiment, the estimated coordinates of a subway inspection vehicle are randomly initialized within a map area. The inspection vehicle, equipped with a positioning module, travels along the sampling path. At each sampling point, the system first selects the six Bluetooth beacons with the strongest Received Signal Strength Indicator (RSSI) as the optimal ranging benchmarks, thus constructing a fixed error objective function. Subsequently, the Adam adaptive moment estimation algorithm is used for iterative optimization, ultimately outputting a high-precision estimated position of the vehicle. Figure 7 As shown in the figure, the convergence curvature diagram of the estimated average position error versus the number of iterations after each iteration update of the 80 sampling points reveals that the green error curve exhibits an extremely excellent convergence trend with increasing iteration count: in the early stages of algorithm iteration, the error drops rapidly with a precipitous drop, demonstrating a very fast convergence rate and effectively overcoming fluctuations caused by signal noise; at 1000 iterations, the model convergence rate reaches a stable minimum, with the average error decreasing to 0.1147 meters; and at 2000 iterations, the final average error remains precisely stable at 0.1147 meters. In summary, this invention can effectively shield against signal interference and ranging anomalies in tunnel environments, achieving centimeter-level high-precision indoor continuous positioning, possessing significant theoretical importance and substantial engineering practical value.

[0050] Secondly, the present invention discloses an adaptive moment estimation optimization positioning system, which includes the above-mentioned adaptive moment estimation optimization positioning method.

[0051] Specifically, the system implements the first aspect of the adaptive moment estimation optimization positioning method. The method mentions that the scheme first deploys Bluetooth beacons on the tunnel wall and records their coordinates, initializes the coordinates of the inspection vehicle, and the vehicle receives the signals of 4-6 of the strongest RSSI Bluetooth beacons in the vicinity. The measured distance with the beacons is calculated through the logarithmic distance path loss model, an optimization objective function is constructed, the gradient of the objective function is calculated and abnormal gradients are clipped, the first-order moment momentum is updated, the second-order moment learning rate is adjusted and the moment estimation deviation is corrected, the vehicle coordinates are iteratively optimized, and finally the gradient clipping and coordinate update are repeatedly performed until the gradient reaches the target or the number of iterations is reached. If it gets stuck in a local optimum, the coordinates are reinitialized and iterated again, and finally the vehicle positioning coordinates are output. By introducing gradient pruning and first-order moment momentum aggregation strategies, the drastic changes in RSSI signals caused by multipath effects in subway tunnels are effectively filtered out, significantly improving positioning stability in complex signal environments. Centimeter-level high-precision positioning can be achieved without additional hardware investment. It has the advantages of low cost, flexible deployment, strong anti-interference, and high robustness. It can stably converge to the global optimum in complex scenarios such as GNSS blind spots in subway tunnels and large initial positioning errors, perfectly adapting to the continuous positioning needs of subway inspection vehicles.

[0052] Thirdly, the present invention discloses an electronic device, which includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When the processor executes the program stored in memory, it implements the steps of the adaptive moment estimation optimization positioning method described above.

[0053] Specifically, the processor runs the computer program in the memory to implement the adaptive moment estimation optimization positioning method in the first aspect. The method mentions that the scheme first deploys Bluetooth beacons on the tunnel wall and records their coordinates, initializes the coordinates of the inspection vehicle, and the vehicle receives the signals of 4-6 of the strongest RSSI Bluetooth beacons in the vicinity. The measured distance with the beacons is calculated through the logarithmic distance path loss model, an optimization objective function is constructed, the gradient of the objective function is calculated and abnormal gradients are clipped, the first-order moment momentum is updated, the second-order moment learning rate is adjusted and the moment estimation deviation is corrected, the vehicle coordinates are iteratively optimized, and finally the gradient clipping and coordinate update are repeatedly executed until the gradient reaches the target or the number of iterations is full. If it gets stuck in a local optimum, the coordinates are reinitialized and iterated again, and finally the vehicle positioning coordinates are output. By introducing gradient pruning and first-order moment momentum aggregation strategies, the drastic changes in RSSI signals caused by multipath effects in subway tunnels are effectively filtered out, significantly improving positioning stability in complex signal environments. Centimeter-level high-precision positioning can be achieved without additional hardware investment. It has the advantages of low cost, flexible deployment, strong anti-interference, and high robustness. It can stably converge to the global optimum in complex scenarios such as GNSS blind spots in subway tunnels and large initial positioning errors, perfectly adapting to the continuous positioning needs of subway inspection vehicles.

[0054] Fourthly, the present invention discloses a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the adaptive moment estimation optimization positioning method as described above.

[0055] Specifically, the computer program on the storage medium is run by a computer to implement the adaptive moment estimation optimization positioning method of the first aspect. As mentioned in the method, this scheme first deploys Bluetooth beacons on the tunnel wall and records their coordinates, initializes the coordinates of the inspection vehicle, and the vehicle receives the signals of 4-6 of the strongest RSSI Bluetooth beacons in the vicinity. The measured distance with the beacons is calculated through the logarithmic distance path loss model, an optimization objective function is constructed, the gradient of the objective function is calculated and abnormal gradients are clipped, the first-order moment momentum is updated, the second-order moment learning rate is adjusted and the moment estimation deviation is corrected, the vehicle coordinates are iteratively optimized, and finally the gradient clipping and coordinate update are repeatedly executed until the gradient reaches the target or the number of iterations is reached. If it gets stuck in a local optimum, the coordinates are reinitialized and iterated again, and finally the vehicle positioning coordinates are output. By introducing gradient pruning and first-order moment momentum aggregation strategies, the drastic changes in RSSI signals caused by multipath effects in subway tunnels are effectively filtered out, significantly improving positioning stability in complex signal environments. Centimeter-level high-precision positioning can be achieved without additional hardware investment. It has the advantages of low cost, flexible deployment, strong anti-interference, and high robustness. It can stably converge to the global optimum in complex scenarios such as GNSS blind spots in subway tunnels and large initial positioning errors, perfectly adapting to the continuous positioning needs of subway inspection vehicles.

[0056] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0057] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0058] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0059] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0060] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0061] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. The illustrative expressions of the above terms in this specification should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0062] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Since these modifications and variations fall within the scope of the claims and their equivalents, this invention also intends to include these modifications and variations.

[0063] The above description describes specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An adaptive matrix estimation optimization positioning method, characterized in that, include, The initial parameters are obtained. The vehicle acquires the signal sent by the Bluetooth beacon, selects a preset number of core beacons based on the signal strength, constructs a logarithmic distance path loss model, and calculates the distance between the vehicle and the core beacon by inputting the signal strength of the core beacon to obtain the measured distance. The initial parameters include the tunnel physical parameters, the coordinates of the Bluetooth beacon in the tunnel, and the coordinates of the first vehicle. Based on the coordinates of the first vehicle, the coordinates of the core beacon, and the measured distance, an objective function for the vehicle coordinates and the core coordinates is constructed. Adaptive gradient calculation and adaptive moment update are performed on the objective function to obtain the updated vehicle coordinates, the L2 norm of the gradient vector, and the number of iterations. Repeat the adaptive gradient calculation and adaptive moment update process until the preset gradient threshold exceeds the L2 norm of the gradient vector or the number of iterations exceeds the preset maximum number of iterations, and obtain the optimized car coordinates.

2. The method according to claim 1, characterized in that, To obtain initial parameters, the vehicle acquires signals from Bluetooth beacons, selects a preset number of core beacons based on signal strength, constructs a logarithmic distance path loss model, and calculates the distance between the vehicle and the core beacons by inputting the signal strength of the core beacons, thus obtaining the measured distance. This process includes the following steps: Obtain tunnel parameters and Bluetooth beacon parameters, install Bluetooth beacons at equal intervals in the tunnel, initialize the vehicle coordinates, and obtain the first vehicle coordinates. Tunnel parameters include tunnel length and Bluetooth beacon coordinates. The car receives Bluetooth beacon wireless signals in real time, sorts them based on signal strength, and selects the beacons with the strongest signal strength as the core beacons. A logarithmic distance path loss model is constructed. The signal strength of the core beacon is input, and the distance between the vehicle and the core beacon is calculated to obtain the measured distance.

3. The method according to claim 1, characterized in that, Based on the coordinates of the first vehicle, the coordinates of the core beacon, and the measured distance, an objective function for the vehicle coordinates and the core coordinates is constructed. Adaptive gradient calculation and adaptive moment update processing are then performed on the objective function to obtain the updated vehicle coordinates, the L2 norm of the gradient vector, and the number of iterations. Specifically, the steps include: Based on the coordinates of the first vehicle, the coordinates of the core beacon, and the measured distance, the theoretical geometric distance between the coordinates of the first vehicle and the coordinates of the core beacon is calculated, and the formula for the theoretical geometric distance is obtained. By combining the theoretical geometric distance and the measured distance, the ranging residual is calculated, and an objective function for positioning optimization is constructed based on the ranging residual, thus obtaining the objective function; The objective function performs adaptive gradient calculation on the coordinates of the first car to obtain the gradient parameters of the objective function. The gradient parameters include the adaptive gradient and the L2 norm of the gradient vector. The adaptive moments are updated based on the gradient parameters of the objective function, and the actual coordinates and iteration number of the car are updated to obtain the updated car coordinates and iteration number.

4. The method according to claim 3, characterized in that, The objective function undergoes adaptive gradient calculation with respect to the coordinates of the first vehicle to obtain the gradient parameters of the objective function. The gradient parameters include the adaptive gradient and the L2 norm of the gradient vector. The specific steps include: Based on the objective function constructed from the coordinates of the first vehicle, the gradient of the objective function with respect to the vehicle coordinates is calculated at the current iteration number, thus obtaining the gradient of the objective function; Calculate the L2 norm of the gradient of the objective function to obtain the L2 norm of the gradient vector; A preset clipping threshold is set. The clipping threshold is compared with the L2 norm of the gradient. If the L2 norm of the gradient vector is greater than the clipping threshold, the gradient is clipped to obtain the clipped gradient.

5. The method according to claim 3, characterized in that, The adaptive moments are updated based on the gradient parameters of the objective function, and the actual coordinates and iteration number of the car are updated to obtain the updated car coordinates and iteration number. The specific steps include: Based on the clipped gradient, the first and second moments are updated to obtain the first and second moment parameters. The first-order moment parameters and the second-order moment parameters are corrected for deviation to obtain the corrected first-order moment and the corrected second-order moment. Based on the corrected first and second moments, the car coordinates are iteratively corrected once, and the number of iterations is accumulated once to obtain the updated car coordinates and the number of iterations.

6. The method according to claim 5, characterized in that, Based on the clipped gradient, the first and second moments are updated to obtain the first and second moment parameters. This process includes the following steps: Obtain the first moment obtained from the previous iteration, the pruned gradient in this iteration, and the preset first moment decay factor. Calculate the first moment according to the weighted fusion rule to obtain the first moment parameters. Obtain the first moment obtained from the previous iteration, the gradient after pruning in this iteration, and the preset second moment decay factor. Calculate the squared value of the gradient after pruning in this iteration. Calculate the second moment by combining the first moment obtained from the previous iteration, the squared value of the gradient after pruning in this iteration, and the preset second moment decay factor according to the weighted fusion rule to obtain the second moment parameter.

7. The method according to claim 1, characterized in that, Repeat the adaptive gradient calculation and adaptive moment update process until the preset gradient threshold exceeds the L2 norm of the gradient vector or the number of iterations exceeds the preset maximum number of iterations, to obtain the optimized car coordinates. The specific steps include: The updated car coordinates are iterated repeatedly. During the iteration, the updated car coordinates are input into the objective function to replace the first car coordinates, and the current iteration process is carried out. The objective function repeatedly undergoes adaptive gradient calculation and adaptive moment update processing to iteratively update the car coordinates; When the preset gradient threshold exceeds the L2 norm of the gradient vector or the number of iterations exceeds the preset maximum number of iterations, stop iterating and updating the car coordinates, and output the optimized car coordinates.

8. An adaptive moment estimation optimization positioning system, characterized in that, The adaptive moment estimation optimization positioning method according to any one of claims 1-7.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the adaptive moment estimation optimization localization method according to any one of claims 1-7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the adaptive moment estimation optimization positioning method as described in any one of claims 1-7.

Images