System and method for improving indoor positioning precision through AI large model

Abstract

The invention discloses a system and method for improving indoor positioning precision through an AI large model, and relates to the technical field of wireless communication. The method comprises the following steps of collecting multi-source signal historical data, and performing data division on the preprocessed multi-source signal historical data for making preparations for AI model training; dividing a target indoor space into a plurality of logic partitions, numbering each logic partition and establishing a coordinate mapping relation; continuously collecting corresponding key indexes in each logic partition, and generating an environmental change index ECD according to the key indexes; the technical key points are as follows: a multi-layer model architecture technical scheme is adopted, a mode of combining basic positioning service and accurate positioning in a complex environment is realized, the service quality of most areas is ensured, a positioning result with higher precision can be provided when necessary, the effect of optimizing the resource utilization efficiency is achieved, and the positioning accuracy is improved. The problem that a traditional single model is poor in performance in a complex environment is solved.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and specifically to a system and method for improving indoor positioning accuracy using an AI large model. Background Art

[0002] Wireless communication refers to the technology of transmitting information through electromagnetic waves in free space without using physical wires to connect the transmitter and receiver; it uses different frequency bands of the radio spectrum to transmit various forms of information such as data, voice, and video; in the scenario of using an AI large model to improve indoor positioning accuracy, wireless communication is a key component; it is necessary to collect multi-source wireless signals through wireless technologies such as Wi-Fi, and these wireless signals provide basic data for positioning and can also support operations such as dynamic adjustment and model update; Traditional indoor positioning technologies generally have problems related to data redundancy or insufficiency caused by fixed sampling frequencies, difficulty of a single model in adapting to complex environmental changes, model aging due to lack of a dynamic update mechanism, and slow response of the centralized computing architecture affecting the user experience; for example, in crowded scenarios such as shopping malls, traditional systems still collect data at a fixed frequency because they cannot perceive the actual signal environment changes, resulting in serious positioning drift when the signal fluctuates greatly; another example is during the peak period of exhibitions, due to relying only on a general model trained once and without deploying edge nodes, the system has a high response delay, inaccurate positioning, and frequent deviation or interruption of the user's navigation path, seriously affecting the actual application effect. Summary of the Invention

[0003] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for improving indoor positioning accuracy using an AI large model, comprising the following steps: Collect historical data of multi-source signals, and perform data partitioning on the preprocessed historical data of multi-source signals to prepare for AI model training; wherein, the historical data of multi-source signals includes multi-source wireless signal data and their corresponding location tags; Divide the target indoor space into several logical partitions, number each logical partition and establish a coordinate mapping relationship; continuously collect the key indicators corresponding to each logical partition, generate an environmental change indicator ECD based on the key indicators, and construct a mode positioning function model to divide different operation modes: When it is determined to be Mode A, it indicates a low ECD, and a lightweight model is constructed; When it is determined to be Mode B, it indicates a medium ECD, and a once-trained model is constructed; When it is determined to be Mode C, it indicates a high ECD, and a twice-trained model is constructed; Among them, the selected AI model is trained using the training set obtained by data partitioning, and after adjustment, a first-trained model is obtained; on the basis of the first training mode, an incremental data set is added, and set data processing actions are executed to output a corrected sampling frequency guidance value and a data increment guidance value to complete the guidance action, and the selected AI model is incrementally trained using the newly partitioned training set, and after adjustment, a second-trained model is obtained. The trained target model is deployed to the server, the user device sends wireless signal data of the current location to the server, and the server uses the target model to obtain the final location and feedback the result.

[0004] Furthermore, the multi-source wireless signal data at least includes: Wi-Fi, Bluetooth, and UWB. Preprocessing the multi-source signal historical data is data cleaning. The multi-source signal historical data is partitioned to obtain a training set and a test set.

[0005] Furthermore, the key indicators at least include: the maximum personnel density and the signal strength change rate. Among them, the maximum personnel density is the maximum number of personnel in the logical partition obtained through the heat map technology within the set time window; the signal strength change rate is calculated by sliding window for RSSI to obtain the signal fluctuation situation per unit time, that is, the signal strength change rate. The process of generating the environmental change index ECD based on the key indicators is as follows: For each logical partition, the environmental change index ECD is generated by weighted summation.

[0006] Furthermore, when constructing the mode positioning function model, the function is as follows: ; In the formula, θ1 and θ2 are respectively two environmental change degree thresholds, and both θ1 and θ2 are greater than 0. ModeA, ModeB, and ModeC are respectively Mode A, Mode B, and Mode C.

[0007] Furthermore, when constructing the lightweight model, knowledge distillation is used. The first-trained model is used as the teacher model, and a student model with 1 / 5 of the number of parameters of the teacher model is trained to provide positioning services for logical partitions with low ECD.

[0008] Furthermore, the incremental data set at least includes: New location tags: Obtain additional location tags in the logical partition with high ECD; Enhance signal data: Adjust the sampling frequency according to the corrected sampling frequency guidance value to capture detailed information, including the signal strength within the environmental change indicator ECD; User behavior data: Record the user's movement trajectory and stay time information.

[0009] Furthermore, the content of the data processing actions is as follows: Data cleaning and data format conversion. With the goal of dynamically adjusting the sampling frequency according to the environmental change indicator ECD, construct an adjustment function model based on the environmental change indicator ECD. Input the environmental change indicator ECD under the current logical partition and output the sampling frequency guidance value. When dealing with the logical partition with high ECD, adopt an adaptive data acquisition strategy. Generate a data increment guidance value based on the difference between the environmental change indicator and the environmental change degree threshold, and use the data increment guidance value as a correction value to feedback-correct the sampling frequency guidance value to obtain the corrected sampling frequency guidance value.

[0010] Furthermore, during the process of deploying the trained target model to the server, it also includes: running a dynamic scheduling mechanism based on spatial neighborhood analysis, including the following steps: Select any logical partition with high ECD as the core partition. With any core partition as the center, the adjacent at least 3 logical partitions are used as the domain area. When there are more than S logical partitions with high ECD in the domain area, trigger the allocation scheduling mechanism to allocate the positioning calculation tasks in the domain area from the original server to the nearest edge AI device; where the value of S is greater than 0.

[0011] A system for an AI large model to improve indoor positioning accuracy, the system includes: Data collection and preprocessing module: Collect multi-source signal historical data, and perform data partitioning on the preprocessed multi-source signal historical data for preparing for AI model training; among them, the multi-source signal historical data includes multi-source wireless signal data and their corresponding location tags; AI large model training module: Divide the target indoor space into several logical partitions, number each logical partition and establish a coordinate mapping relationship; Continuously collect the key indicators corresponding to each logical partition, generate the environmental change indicator ECD based on the key indicators, and construct a pattern positioning function model to divide different operation modes: When it is determined as mode A, it means low ECD, and construct a lightweight model; When it is determined as mode B, it means medium ECD, and construct a once-trained model; When it is determined as mode C, it means high ECD, and construct a twice-trained model; Among them, the selected AI model is trained using the training set obtained by data partitioning, and after adjustment, a primary training model is obtained; an incremental data set is added on the basis of the primary training mode, and set data processing actions are executed to output a corrected sampling frequency guidance value and a data increment guidance value to complete the guidance action, and the selected AI model is incrementally trained using the newly partitioned training set, and after adjustment, a secondary training model is obtained; Location service deployment module: Deploy the trained target model to the server. The user device sends wireless signal data of the current location to the server, and the server uses the target model to obtain the final location and feedback the result.

[0012] The present invention provides a system and method for improving indoor positioning accuracy by an AI large model, which has the following beneficial effects: (1) This solution adopts a dynamic scheduling mechanism of a primary training model (accompanied by a lightweight model) + a secondary training model + spatial neighborhood analysis, that is, a technical solution of a multi-level model architecture, realizing a combination of basic positioning services and precise positioning in complex environments. It not only ensures the service quality in most areas but also can provide higher-precision positioning results when necessary, achieving the effect of optimizing resource utilization efficiency and solving the problem that traditional single models perform poorly in complex environments; in addition, combined with the dynamic scheduling mechanism, it further improves the system response speed and service quality and reduces the pressure on the central server; (2) Through region partitioning and environmental change detection, this solution can achieve targeted modeling of high-frequency interference regions, solving the problem of interference of dynamic environments on signal propagation; introducing region-specific models to avoid overfitting or underfitting of the main model due to mixed training data, improving the generalization ability and robustness of the model: only enabling higher-precision models in necessary regions to avoid high-cost calculations in the entire region, taking into account both performance and resources, and achieving a dynamic balance between positioning accuracy and efficiency; through real-time monitoring and dynamic training mechanisms, the system has the ability to quickly adapt to sudden situations such as peak traffic and structural changes, supporting rapid response to sudden environmental changes; (3) By adopting the technical solution of real-time monitoring of the ECD value, it realizes precise perception of complex environmental changes, and dynamically adjusts the sampling frequency and data volume according to the ECD value to ensure sufficient information is obtained in high-fluctuation regions; achieving the effect of improving data quality and model accuracy, while solving the problem of data insufficiency or waste caused by fixed sampling frequency in traditional technologies. This linkage mechanism ensures that the system can adaptively respond to different environmental conditions and improves the overall stability of the positioning service; (4)Adopting the technical solution of an adaptive data acquisition strategy, the goal of automatically adjusting the data acquisition volume according to the degree of environmental change is achieved, ensuring high-quality data required for model update, achieving the effect of maintaining the long-term effectiveness and robustness of the model, and solving the problem of model obsolescence or inaccuracy caused by lack of timely update in traditional methods. This process not only enhances the adaptability of the model but also reduces unnecessary resource consumption; (5)Compared with migrating tasks every time a single high-ECD area is detected, this solution reduces unnecessary scheduling times and system overhead, thereby improving efficiency; due to more precise allocation of computing resources, users can also obtain continuous and high-quality positioning services in complex environments, reducing service interruptions or delays caused by frequent switching, achieving a better user experience; the method based on spatial neighborhood analysis provides a more refined and effective resource management method, ensuring service quality while avoiding unnecessary resource waste. Brief Description of the Drawings

[0013] Figure 1 It is a flowchart of a method for an AI large model to improve indoor positioning accuracy according to the present invention. Detailed Embodiments

[0014] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0015] Embodiment 1: Please refer to Figure 1 , this embodiment provides a method for an AI large model to improve indoor positioning accuracy. The solution recorded in this method aims to improve the accuracy of indoor positioning by combining an AI large model to optimize the recognition and analysis of the propagation characteristics of wireless signals in a complex indoor environment; the specific steps of this method are as follows: S1. Data collection and preprocessing: S1.1. Collect various types of wireless signal data (i.e., multi-source wireless signal data) and corresponding location tags in different indoor environments; among them, the multi-source wireless signal data includes at least: Wi-Fi, Bluetooth, and UWB; S1.2. Clean the collected data to remove noise and outliers; S1.3. Divide the data into a training set and a test set to prepare for subsequent AI model training; among them, the ratio of the training set to the test set can be selected as 7:3, and it can be specifically selected according to actual needs.

[0016] S2. AI large model training: S2.1. First training model: Select a suitable AI model architecture to process multi-source wireless signal data; among them, the AI model architecture includes at least any one of the deep neural network DNN or the convolutional neural network CNN; use the training set to train the selected AI model, and adjust the model parameters to minimize the error between the predicted position and the actual position to obtain the first training model.

[0017] S2.2. Lightweight model: Adopt knowledge distillation, use the first training model as the teacher model, and train a student model with 1 / 5 of the number of parameters of the teacher model for rapid positioning service in the low ECD area; among them, the lightweight model is derived from the first training model and is not a brand-new model independently trained; it can be regarded as a "miniaturized version of the first model" for quickly responding to positioning requests in a simple environment; In addition, in addition to the above method of using knowledge distillation, model pruning can also be selected; use a neural network structure with fewer layers and smaller number of parameters; or only retain the feature extraction layer in the main model that has a greater impact on a specific area, so as to realize the simplification operation of the first training model. Other simplification methods are existing technologies, so they will not be elaborated here.

[0018] S2.3. Second training model: S2.3.1. Region division and feature collection: Divide the target indoor space into several logical partitions (such as floors, rooms, corridors, etc.), number each area and establish a coordinate mapping relationship; deploy index devices in each logical partition to continuously collect key indicators; among them, the index devices include at least: wireless access points (APs), Bluetooth beacons, and UWB base stations; The key indicators include the following two categories: Maximum personnel density: Obtain the maximum number of people in the area within a set time window (such as 10 seconds) through Wi-Fi probe requests, camera recognition, or heat map technology; Signal strength change rate: Perform a sliding window calculation on RSSI (Received Signal Strength Indicator) to obtain the signal fluctuation situation per unit time for evaluating signal stability; S2.3.2. Analysis of environmental change degree: For each logical partition, use the method of weighted summation to generate the corresponding environmental change index under each logical partition, and its formula is as follows: ; In the formula, ECD represents the environmental change index, reflecting the fluctuation degree of the wireless signal environment under the current logical partition; w1 and w2 are weight coefficients respectively, and their value ranges are both [0, 1], which can be adjusted according to the actual scenario requirements and are used to balance the influence of personnel density and signal change; rd_max and R_max represent the maximum personnel density and the upper limit of the regional capacity (the maximum number of people that can be accommodated in each preset area) respectively; sr and sr_th represent the signal strength change rate (the standard deviation of the signal strength per unit time) and the standard fluctuation threshold (a preset reference value, representing the signal strength fluctuation value under normal circumstances) respectively; S2.3.3. Adjust the positioning service mode: Construct the following mode positioning function model to divide different operation modes: ; In the formula, θ1 and θ2 represent two environmental change degree thresholds respectively, and both θ1 and θ2 are greater than 0; θ1 < θ2; Two environmental change degree thresholds are used to divide different operation modes: ModeA: Low ECD, adopt a lightweight model, emphasizing fast response; ModeB: Medium ECD, use a once-trained model (i.e., the standard model), balancing accuracy and efficiency; ModeC: High ECD, enable a twice-trained model optimized for complex environments, focusing on robustness and accuracy; The actual effects in different modes are illustrated in Table 1 below: Table 1: Reference for operation effect indicators in different modes: Index Mode A (Lightweight) Mode B (Standard) Mode C (Second Training) Positioning Accuracy ±2.0m ±1.5m ±1.0m Inference Time < 100ms 0 - 150ms 0 - 200ms Resource Consumption Low Medium High Applicable Scenario Stable Low - interference Area Normal Area Dynamic Complex Area S2.3.4. Incremental training of the twice-trained model: Collect the original dataset and the incremental dataset, perform the set data processing actions, output the corrected sampling frequency guidance value and the data increment guidance value, and adjust the sampling frequency according to the corrected sampling frequency guidance value, and determine the total acquisition volume of the original dataset and the incremental dataset according to the data increment guidance value; Then re-divide them into a training set and a test set, and perform incremental training on the selected AI model using the training set, and adjust the model parameters to minimize the error between the predicted position and the actual position to obtain the twice-trained model; By adopting the above technical solutions, a multi-level model architecture and intelligent scheduling are achieved: Through a dynamic scheduling mechanism that combines a one-time trained model (accompanied by a lightweight model), a second-time trained model, and spatial neighborhood analysis, a combined technical effect is achieved. Specifically: By adopting a technical solution with a multi-level model architecture, a method of combining basic positioning services with precise positioning in complex environments is realized. This not only ensures the service quality in most areas but also provides higher-precision positioning results when necessary, achieving the effect of optimizing resource utilization efficiency, solving the problem that traditional single models perform poorly in complex environments. Additionally, combined with the dynamic scheduling mechanism, the system response speed and service quality are further improved, and the pressure on the central server is reduced.

[0019] It should be noted that a two-layer AI model architecture is deployed: The first layer is a general main model (one-time trained model), which is applicable to basic positioning in all areas; The second layer is a region-specific model (second-time trained model), which is only used when the corresponding logical partition is activated; For the one-time and second-time trained models, the following specific scenario descriptions are given: In a large convention and exhibition center, there are multiple exhibition halls, rest areas, and entrances and exits. Usually, the number of people in each area is relatively stable, but during the peak period of the exhibition, there will be a phenomenon of dense people in some popular exhibition areas; The application process is as follows: Region division: The convention and exhibition center is divided into five exhibition areas A1 - A5, two rest areas B1 - B2, and three entrance and exit channels C1 - C3; Environmental monitoring: After the exhibition starts, due to the exhibition content in Area A3 attracting a large number of audiences, the personnel density reaches more than 80% within 10 seconds, and the Wi-Fi signal fluctuates violently (the change rate exceeds the standard value); Trigger mechanism: It is judged that the environmental change degree in Area A3 exceeds the standard, and data enhancement collection is immediately started, the sampling frequency is increased, and multiple groups of data such as signal fingerprints under high density are recorded; Second-time training: The model in Area A3 is incrementally trained using the newly added data to generate a positioning model specifically for the high-density crowd scenario; Positioning service: When a user enters Area A3, the system first determines the position in Area A3 by the main model, and then automatically switches to the A3 exclusive model for more accurate positioning output.

[0020] The actual effect is illustrated through Table 2 below: Table 2: Comparison of effect indicators after the actual operation of the traditional method and this solution: Index Traditional Method This Solution Positioning Accuracy (m) ±2.0~3.0 ±1.0 - 1.5 (After Optimization in Dynamic Area) Response Delay (ms) < 200ms < 250ms (Including Model Switching) Adaptability to Complex Environments Poor (Prone to Interference) Significantly Improved (Dynamic Adaptation) Data Update Frequency Fixed Period Real - time / On - demand It can be seen from the above table content that: Through regional division and environmental change detection, this solution can achieve targeted modeling of high-frequency interference areas, solve the problem of interference of dynamic environment on signal propagation; introduce regional dedicated models to avoid overfitting or underfitting of the main model due to mixed training data, and improve the generalization ability and robustness of the model: only enable higher-precision models in necessary areas to avoid high-overhead calculations in the entire area, balance performance and resources, and achieve a dynamic balance between positioning accuracy and efficiency; through real-time monitoring and dynamic training mechanisms, the system can quickly adapt to sudden situations such as peak human flow and structural changes, and support rapid response to sudden environmental changes.

[0021] Among them, the original data set: includes the initial data used for the first training of the model (i.e., the multi-source wireless signal data in S1), which cover multiple regions and reflect the signal characteristics under different environmental conditions; The incremental data set includes at least: New location tags: In areas with high ECD, obtain additional location tags by deploying more positioning devices or using more precise methods (the amount of additional location tags can be set according to the area of the actual region, and the two are positively correlated); enhanced signal data: adjust the sampling frequency (for example, from once per second to once every 0.5 seconds) to capture more detailed information, including the signal strength within the environmental change index, especially in the case of large signal fluctuations; user behavior data: record the movement trajectory and stay time information of users to help understand the human flow pattern in the environment.

[0022] By adopting the above technical solutions, accurate environmental perception and dynamic adjustment are achieved; Through real-time monitoring of the ECD value + dynamic adjustment of the sampling frequency, a combined technical effect is achieved. Specifically: by adopting the technical solution of real-time monitoring of the ECD value, accurate perception of complex environmental changes is realized, and the sampling frequency and data volume are dynamically adjusted according to the ECD value to ensure sufficient information is obtained in high-fluctuation areas; the effect of improving data quality and model accuracy is achieved, and at the same time, the problem of insufficient or wasted data caused by fixed sampling frequency in traditional technologies is solved. This linkage mechanism ensures that the system can adaptively respond to different environmental conditions and improves the overall stability of the positioning service.

[0023] The content of the data processing actions is as follows: Data cleaning and data format conversion, aiming at dynamically adjusting the sampling frequency according to the environmental change index ECD, constructing an adjustment function model based on the environmental change index ECD, inputting the environmental change index ECD under the current logical partition, and outputting the sampling frequency guidance value; For the logical partition with high ECD, an adaptive data acquisition strategy is adopted. Based on the difference between the environmental change index and the environmental change degree threshold, a data increment guidance value is generated, and the data increment guidance value is used as a correction value to feedback and correct the sampling frequency guidance value, and the corrected sampling frequency guidance value is obtained; The above operations are performed according to the corrected sampling frequency guidance value when adjusting the sampling frequency; During the incremental training of the secondary training model in S2.3.4, at least the data increment guidance value (for example: 600) data points need to be collected in the logical partition corresponding to the high ECD for training; Among them, data cleaning is to remove noise and outliers; Data format conversion is to ensure that all data is converted into a format suitable for model input; The function running in the adjustment function model is defined as: ; In the formula, f sample represents the output sampling frequency guidance value (unit: Hz), f min and f max represent the minimum sampling frequency and the maximum sampling frequency respectively. For example: the minimum sampling frequency and the maximum sampling frequency are 1 Hz (once per second) and 2 Hz (once every 0.5 seconds) respectively; When the obtained result f sample exceeds the set upper limit f max = 2.0, according to the above setting, it is forced to be set as the maximum sampling frequency.

[0024] The logical explanation of the above function is: When the ECD is small (close to θ1), the sampling frequency is close to f min , that is, low-frequency acquisition; as the ECD rises, the sampling frequency increases linearly; when the ECD reaches or exceeds θ2, the sampling frequency reaches f max , entering the high-frequency acquisition state; adopting the above adjustment function model design enables the system to allocate resources on demand, avoiding unnecessary high-frequency acquisition from wasting computing power, and improving data quality when needed.

[0025] The method of generating the data increment guidance value is as follows: ; In the formula, D represents the required data increment guidance value under the corresponding logical partition, D baseD represents the basic data volume, that is, the minimum required data acquisition volume, ∆θ represents the additional data volume coefficient corresponding to each unit by which the ECD exceeds the environmental change degree threshold θ2, and its value range is [0, 1], usually taking the value of 1. K represents a proportionality factor used to adjust the growth rate of the data volume, and its value range is greater than 0, which is set according to actual requirements; It should be noted that the above formula linearly increases the required data volume according to the degree to which the ECD value exceeds the threshold; the basic data volume D base ensures the minimum acquisition requirement, and the excess part is expanded according to the proportionality factor K and the unit increment coefficient ∆θ, so that the data acquisition volume adapts to the growth of the environmental complexity, and the adaptability of the model is improved.

[0026] When obtaining the corrected sampling frequency guidance value, that is, the function running in the corrected adjustment function model is as follows: ; In the formula, fsample_x represents the corrected sampling frequency guidance value, γ represents the data volume weight coefficient, and its value range is [0, 1], which controls the influence intensity of the data volume on the frequency; The logical description is: the original part represents the basic adjustment based on the ECD; New item As a correction factor, when D > Dbase, the sampling frequency is increased to accelerate the data collection speed; γ controls the strength of the correction amplification effect to prevent the frequency from mutating too much; By introducing the data increment guidance value D to dynamically correct the sampling frequency, it is possible to automatically increase the sampling density in the case of high ECD and large data requirements, ensure sufficient model training data, and avoid resource waste at the same time.

[0027] In S2, optionally, the following steps are further included: Evaluate the model performance on the test set and further optimize the selected AI model according to the results; Specifically, a series of standard metrics are used to measure the model performance (i.e., performance) when evaluating the model performance, such as positioning error (mean absolute error MAE, root mean square error RMSE), accuracy, recall rate, etc.; analyze the results: if the positioning error is high, it indicates that the model performs poorly in some regions or conditions; if the accuracy is low and the recall rate is high, it may mean that the model is too conservative; otherwise, it may be too aggressive; the specific methods for optimizing the model at least include: hyperparameter tuning: adjusting the hyperparameters of the model, such as learning rate, regularization coefficient, etc., through grid search or random search to find the best configuration; model structure optimization: simplifying complex models to reduce overfitting, or increasing the number of layers / nodes to improve the model's expression ability.

[0028] By adopting the above technical solution, adaptive data collection and model update are achieved: Adopting the technical solution of the adaptive data collection strategy realizes the goal of automatically adjusting the data collection volume according to the degree of environmental change, ensures the high-quality data required for model update, achieves the effect of maintaining the long-term effectiveness and robustness of the model, and solves the problem of model obsolescence or inaccuracy caused by the lack of timely update in traditional methods. This process not only enhances the adaptability of the model but also reduces unnecessary resource consumption.

[0029] Embodiment 2: Based on Embodiment 1, this embodiment also provides S3, positioning service deployment: S3.1 Deploy the trained model to the server to process the wireless signal data from user devices in real time; among them, it also includes: a dynamic scheduling mechanism based on spatial neighborhood analysis: Real-time screen out any logical partition with a high ECD as the core partition. Taking any core partition as the center, the adjacent at least 3 logical partitions are used as the domain area (assuming each logical partition is a square or rectangular grid unit, then a domain area has at least 7×7 logical partitions); when there are more than S logical partitions with a high ECD in the domain area, trigger the allocation scheduling mechanism, and allocate the positioning calculation tasks in the domain area from the original server to the nearest edge AI device. (The edge AI device uses the locally cached data for quick response and dynamically adjusts the resource allocation according to the actual load situation to ensure service quality and user experience; among them, how to dynamically adjust the allocation is an existing technology and has nothing to do with the core technical points in this solution, so it will not be elaborated here); The value of S is greater than 0 and is set according to actual needs, and can be set to 3; The logic code of the allocation scheduling mechanism is as follows: def should_dispatch_to_edge(region, ecd_map, threshold=1.8, min_neighbors=3): # Obtain the ECD value of the current logical partition current_ecd = ecd_map[region] if current_ecd<threshold: return False # Define the neighborhood range neighbors = get_neighbors(region) # Count the number of neighbors that meet the conditions high_ecd_neighbors = sum(ecd_map[neighbor] >= threshold for neighbor in neighbors) # Determine whether to schedule to the edge computing node if high_ecd_neighbors >= min_neighbors: return True else: return False; Where: region represents the core partition (area) to be evaluated; ecd_map represents a dictionary or mapping table that stores all regions and their corresponding Environmental Change Metrics (ECD); The get_neighbors(region) function returns a list of all directly adjacent regions of the given region; threshold is the threshold of ECD, that is, θ2, which can be default set to 1.8 according to requirements; min_neighbors is the required number of high-ECD neighbors, default set to 3; It should be noted that the XX partition and XX region mentioned in this embodiment have the same meaning; Specifically, this embodiment can verify the effectiveness of this mechanism by comparing the average response time (ms) and the mean positioning error (meter) under different scheduling mechanisms, as shown in Table 3 below: Table 3: Effect data reference under different scheduling mechanisms: Scheduling Mechanism Average Response Time (ms) Positioning Error RMSE (m) Remarks (Central) Server 250 ms ±1.5 m May be Delayed in High - fluctuation Areas Edge Computing (AI) Device 150 ms ±1.2 m Fast Response, Reduced Error As can be seen from Table 3 above, after adopting edge computing (AI) devices, not only the average response time is significantly reduced, but also the positioning error is further reduced, especially in the case of large signal fluctuations, the effect is particularly obvious; S3.2. The user device sends the wireless signal information of the current location to the server through the APP or other forms; S3.3. The server uses the AI model to calculate the most likely location (i.e., the final location) and feedback the result to the user device.

[0030] By combining the spatial distribution characteristics of ECD values with the dynamic scheduling mechanism, this solution can optimize the resource utilization efficiency while ensuring the service quality, and is particularly suitable for scenarios with dense crowds and changing environments such as shopping malls, airports, and hospitals; it not only improves the robustness and adaptability of the system, but also enhances the user experience, making the positioning service more intelligent and efficient.

[0031] Generally, the conventional solution adopted by those skilled in the art when considering dynamic scheduling is as follows: Immediately initiate scheduling the positioning calculation task from the central server to the edge node (Edge AI) close to the user in a single area with a high ECD value, which is an intuitive and conventional solution; once it is detected that the ECD value of a certain area exceeds the threshold, immediately transferring the tasks in that area to the edge node can quickly reduce the pressure on the central server and may improve the local service quality; however, the above solution has certain limitations: Resource waste: If there is only a short-term high ECD value fluctuation in a single area, and this fluctuation is not significant enough to affect the overall service, immediately mobilizing edge computing resources may lead to unnecessary resource consumption; Load balancing problem: Frequent switching of calculation tasks between individual high ECD areas may lead to uneven load among edge nodes, with some nodes being overloaded while others are idle.

[0032] Insufficient adaptability to complex environments: Making decisions based only on the ECD value of a single area may ignore the influence of surrounding areas; for example, if multiple adjacent areas simultaneously experience high ECD values, there may be mutual interference or common problem sources among these areas, which need to be handled uniformly; Therefore, the reasons and advantages for adopting the dynamic scheduling mechanism based on spatial neighborhood analysis in this embodiment are as follows: Reasons: By considering the ECD value situation within and around a region, it is possible to more accurately determine whether there are extensive environmental disturbances; this helps to identify those challenges that are not just local problems but systemic challenges existing within a larger scope; only when it is confirmed that there is a widespread impact (i.e., at least S adjacent areas also show high ECD values) is edge computing triggered, which can ensure that resources are used where they are truly needed and avoid unnecessary migration costs; in the case where multiple regions are jointly affected, centrally utilizing edge computing capabilities can better handle complex signal propagation conditions and provide a more stable and reliable positioning service; Advantages: Compared with migrating tasks every time a single high ECD area is detected, this method reduces unnecessary scheduling times, lowers the system overhead, and thus improves efficiency; due to more precise allocation of computing resources, users can also obtain continuous and high-quality positioning services in complex environments, reducing service interruptions or delays caused by frequent switching and achieving a better user experience; by uniformly managing multiple associated high ECD areas, it can provide data support for future network layout optimization, device deployment, etc., which helps to formulate long-term strategic plans; In summary, although the conventional solution is simpler and more straightforward, the method based on spatial neighborhood analysis provides a more refined and effective resource management approach, which not only ensures the quality of service but also avoids unnecessary resource waste; this is crucial for the subsequent construction of an efficient and stable indoor positioning system.

[0033] Embodiment 3: Based on Embodiment 1 and Embodiment 2, this embodiment also provides a system for improving indoor positioning accuracy using an AI large model. The system includes: Data collection and preprocessing module: Collect multi-source signal historical data, and perform data partitioning on the preprocessed multi-source signal historical data to prepare for AI model training; among them, the multi-source signal historical data includes multi-source wireless signal data and their corresponding location tags. AI large model training module: Divide the target indoor space into several logical partitions, number each logical partition, and establish a coordinate mapping relationship; continuously collect the key indicators corresponding to each logical partition, generate an environmental change indicator ECD based on the key indicators, and construct a pattern positioning function model to divide different operation modes: When it is determined as Mode A, it indicates a low ECD, and a lightweight model is constructed; When it is determined as Mode B, it indicates a medium ECD, and a one-time training model is constructed; When it is determined as Mode C, it indicates a high ECD, and a two-time training model is constructed; Among them, the selected AI model is trained using the training set obtained by data partitioning, and after adjustment, a one-time training model is obtained; on the basis of the one-time training mode, an incremental data set is added, the set data processing actions are executed, the corrected sampling frequency guidance value and data increment guidance value are output to complete the guidance action, and the selected AI model is incrementally trained using the newly partitioned training set, and after adjustment, a two-time training model is obtained; Positioning service deployment module: Deploy the trained target model to the server. The user device sends the wireless signal data of the current location to the server, and the server uses the target model to obtain the final location and feedback the result.

[0034] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or in the combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution.

[0035] The unit described as a separation component may or may not be physically separated. The component displayed as a unit may or may not be a physical unit, and it may be located in one place or distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0036] As described above, the foregoing are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. A method for improving indoor positioning accuracy by an AI large model, characterized in that, It includes the following steps: Collect historical data of multi-source signals, and perform data partitioning on the preprocessed historical data of multi-source signals to prepare for AI model training; among them, the historical data of multi-source signals includes multi-source wireless signal data and its corresponding location tags; Divide the target indoor space into several logical partitions, number each logical partition and establish a coordinate mapping relationship; continuously collect the corresponding key indicators in each logical partition, generate an environmental change indicator ECD based on the key indicators, and construct a mode positioning function model to divide different operation modes: When it is determined to be Mode A, it indicates a low ECD, and a lightweight model is constructed; When it is determined to be Mode B, it indicates a medium ECD, and a one-time training model is constructed; When it is determined to be Mode C, it indicates a high ECD, and a secondary training model is constructed; Among them, use the training set obtained by data partitioning to train the selected AI model, and adjust it to obtain a one-time training model; add an incremental data set on the basis of the one-time training mode, execute the set data processing actions, output the corrected sampling frequency guidance value and data increment guidance value to complete the guidance action, and use the newly partitioned training set to perform incremental training on the selected AI model, and adjust it to obtain a secondary training model; Deploy the trained target model to the server, the user device sends the wireless signal data of the current location to the server, and the server uses the target model to obtain the final location and feedback the result.

2. The method for improving indoor positioning accuracy of an AI large model according to claim 1, characterized in that: The multi-source wireless signal data at least includes: Wi-Fi, Bluetooth, and UWB; The preprocessing of the historical data of multi-source signals is data cleaning; Perform data partitioning on the historical data of multi-source signals to obtain a training set and a test set.

3. A method for improving indoor positioning accuracy using an AI large model according to claim 1, characterized in that: The key indicators at least include: the maximum value of personnel density and the signal strength change rate; Among them, the maximum value of personnel density is obtained by using the heat map technology to obtain the maximum number of people in the logical partition within the set time window; the signal strength change rate is calculated by a sliding window of RSSI to obtain the signal fluctuation situation per unit time, that is, the signal strength change rate; The process of generating the environmental change indicator ECD based on the key indicators is: For each logical partition, the environmental change indicator ECD is generated by using the weighted summation method.

4. A method for improving indoor positioning accuracy of an AI large model according to claim 1, characterized in that: When constructing the mode positioning function model, the function is as follows: ； In the formula, θ1 and θ2 are two environmental change degree thresholds respectively, and both θ1 and θ2 are greater than 0; ModeA, ModeB, and ModeC are Mode A, Mode B, and Mode C respectively.

5. The method for improving indoor positioning accuracy of an AI large model according to claim 1, characterized in that: When constructing the lightweight model, knowledge distillation is used, and the one-time training model is used as the teacher model to train a student model with a parameter quantity of 1 / 5 of the teacher model for positioning services in logical partitions with low ECD.

6. The method for improving indoor positioning accuracy by an AI large model according to claim 1, characterized in that: The incremental data set at least includes: Newly added location tags: In the logical partition with high ECD, obtain additional location tags; enhanced signal data: Adjust the sampling frequency according to the corrected sampling frequency guidance value to capture detailed information, including the signal strength within the environmental change indicator ECD; user behavior data: Record the movement trajectory and stay time information of the user.

7. A method for improving indoor positioning accuracy by an AI large model according to claim 6, characterized in that: The content of the data processing actions is as follows: Data cleaning and data format conversion, aiming at dynamically adjusting the sampling frequency according to the environmental change index ECD, constructing an adjustment function model based on the environmental change index ECD, inputting the environmental change index ECD under the current logical partition, and outputting the sampling frequency guidance value; when dealing with logical partitions with high ECD, an adaptive data acquisition strategy is adopted, generating a data increment guidance value based on the difference between the environmental change index and the environmental change degree threshold, and using the data increment guidance value as a correction value to feedback and correct the sampling frequency guidance value to obtain the corrected sampling frequency guidance value.

8. A method for improving indoor positioning accuracy of an AI large model according to claim 1, characterized in that: During the process of deploying the trained target model to the server, it also includes: running a dynamic scheduling mechanism based on spatial neighborhood analysis, including the following steps: Select any logical partition with high ECD as the core partition, and take the adjacent at least 3 logical partitions within the area centered on any core partition as the domain area; when there are more than S logical partitions with high ECD in the domain area, trigger the allocation scheduling mechanism to allocate the positioning calculation tasks in the domain area from the original server to the nearest edge AI device; where the value of S is greater than 0.

9. A system for improving indoor positioning accuracy by an AI large model, characterized in that: The system includes: Data collection and preprocessing module: collect multi-source signal historical data, and perform data partitioning on the preprocessed multi-source signal historical data to prepare for AI model training; among them, the multi-source signal historical data includes multi-source wireless signal data and its corresponding location tags. AI large model training module: divide the target indoor space into several logical partitions, number each logical partition and establish a coordinate mapping relationship; continuously collect the key indicators corresponding to each logical partition, generate the environmental change index ECD based on the key indicators, and construct a pattern positioning function model to divide different operation modes: When it is determined to be mode A, it means low ECD, and a lightweight model is constructed; When it is determined to be mode B, it means medium ECD, and a once-trained model is constructed; When it is determined to be mode C, it means high ECD, and a twice-trained model is constructed; Among them, use the training set obtained by data partitioning to train the selected AI model, and after adjustment to obtain a once-trained model; add an incremental data set on the basis of the once-training mode, execute the set data processing actions, output the corrected sampling frequency guidance value and data increment guidance value to complete the guidance action, and use the newly partitioned training set to perform incremental training on the selected AI model, and after adjustment to obtain a twice-trained model; Positioning service deployment module: deploy the trained target model to the server, the user device sends the wireless signal data of the current location to the server, and the server uses the target model to obtain the final location and feedback the result.

Images