Asset positioning monitoring method based on Wi-Fi and inertial navigation fusion and large language model

Abstract

The invention belongs to the technical field of industrial asset protection and positioning, and discloses an asset positioning monitoring method based on Wi-Fi and inertial navigation fusion and a large language model, and the method comprises the following steps: obtaining multi-modal data, and constructing a fingerprint database; the industrial assets are accurately positioned through the fusion positioning technology of Wi-Fi and inertial navigation in combination with the acquired signal strength, the inertial navigation path and the feature data in the fingerprint database; and performing semantic fusion and analysis on the multi-modal data by using a large language model, identifying potential risks, and generating a safety assessment report. According to the asset positioning monitoring method based on Wi-Fi and inertial navigation fusion and the large language model, accurate tracking of asset positions can be realized, asset operation states can be sensed and potential risks can be identified through a multi-modal data fusion technology and an anomaly detection algorithm, and real-time risk early warning can be provided.

Description

Technical Field

[0001] The present invention relates to the field of industrial asset protection and positioning technology, and in particular to an asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and a large language model. Background Art

[0002] In modern industrial environments, accurate asset positioning and protection are key to ensuring production efficiency and safety. Traditional industrial asset positioning methods usually rely on a single technical means. These single technical methods each have some limitations and are difficult to cope with complex and dynamically changing industrial environments, especially indoors or in areas with severe signal interference. Positioning accuracy often cannot meet actual needs. Wi-Fi and inertial navigation systems combined with extended Kalman filter fusion can effectively make up for their respective shortcomings and improve positioning accuracy and stability through multi-sensor data fusion. This method can improve positioning accuracy and enhance the system's adaptability to signal fluctuations and sensor errors.

[0003] Traditional industrial asset positioning methods mainly focus on obtaining asset location information, and their core goal is to provide basic asset location services. However, these methods usually rely on a single technical means, have relatively simple functions, and can only meet the basic positioning needs of assets in static or conventional dynamic environments. In complex industrial environments, such methods often fail to meet the high standards of industrial asset protection and safety management. In addition, traditional positioning methods usually lack the ability to fully perceive and analyze asset status, and are unable to promptly identify potential abnormal conditions or provide effective risk warnings.

[0004] In the field of industrial asset protection, the positioning function has gradually changed from "knowing the location" to "understanding the status" and "warning of risks". In the modern industrial environment, the value of assets is not only reflected in their location data, but also in the monitoring of their operating status and the identification and prevention of risks. Therefore, it is urgent to develop a more intelligent positioning and monitoring method for important industrial assets, which can not only accurately track the location of assets, but also perceive the operating status of assets, identify potential risks, and provide real-time risk warnings through multimodal data fusion technology and anomaly detection algorithms. Summary of the invention

[0005] The purpose of the present invention is to provide an asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and large language model, so as to improve the positioning accuracy and have the functions of anomaly detection and risk warning, so as to realize comprehensive intelligent protection of industrial assets.

[0006] To achieve the above object, the present invention provides an asset positioning monitoring method based on Wi-Fi, inertial navigation fusion and large language model, comprising the following steps:

[0007] Step S1, obtaining multimodal data and building a fingerprint library; wherein the multimodal data includes Wi-Fi signals, inertial navigation data, device logs and image data;

[0008] Step S2: accurately locate industrial assets by using the fusion positioning technology of Wi-Fi and inertial navigation, combining the collected signal strength, inertial navigation path and feature data in the fingerprint library;

[0009] Step S3: Use a large language model to semantically fuse and analyze multimodal data, identify potential risks, and generate a safety assessment report including safety ratings, key findings, and warning recommendations.

[0010] Preferably, in step S1, multimodal data is obtained to construct a fingerprint library; wherein the multimodal data includes Wi-Fi signals, inertial navigation data, device logs and image data, and the specific process is as follows:

[0011] Step S11: Collect data D from different sources using a multimodal data acquisition device t , as shown below:

[0012]

[0013] in, Indicates Wi-Fi signal data, Indicates inertial navigation data, Indicates device log and image data; the superscript t indicates the sampling time;

[0014] Step S12: Obtain signal strength values ​​D through multiple Wi-Fi access points in the environment WiFi , as shown below:

[0015] D WiFi ={rssi1, rssi2,…, rssi n};

[0016] Among them, D WiFi represents a set of signal strength values ​​from n access points;

[0017] Step S13: Obtain the inertial navigation data of the device through the inertial sensor, including the linear acceleration D Linear , gravity acceleration D Gravity and the rotation angle D Rotation , as shown below:

[0018] D Linear = {linear x , linear y , linear z};

[0019] D Gravity ={gravity x ,gravity y , gravity z};

[0020] D Rotation ={rotation x , rotation y , rotation z , rotation w};

[0021] The set of the above inertial navigation data is defined as:

[0022] D IN ={D Linear , D Gravity , D Rotation};

[0023] Step S14: Obtain operation log information from industrial equipment, including timestamp, current status of equipment, operation records, alarm information, and maintenance history, as shown below:

[0024] D Log ={L1, L2, L3, L4, L5};

[0025] Step S15: Collect real-time image data D through the camera Image ; Among them, D Image Represents the acquired image data.

[0026] Preferably, in step S2, the industrial assets are accurately located by using the fusion positioning technology of Wi-Fi and inertial navigation, combined with the collected signal strength, inertial navigation path and feature data in the fingerprint library. The specific process is as follows:

[0027] Step S21, preprocessing the collected multimodal data including Wi-Fi signal strength and inertial navigation data;

[0028] Step S22: using inertial navigation technology to estimate the relative change of the device position;

[0029] Step S23: During the calculation process, the moving direction and step length of the device are calculated using the output of the inertial sensor, and the position of the device is gradually updated;

[0030] Step S24, using the rotating vector quaternion of the integrated accelerometer, magnetometer and gyroscope data to perform heading calculation;

[0031] Step S25: Based on the previous position (x k ,yk ), step length L k and direction ψ k , calculate the current position of the device (x k+1 ,y k+1 );

[0032] Step S26: combining the real-time Wi-Fi signal strength data and the pre-built fingerprint database, using a convolutional neural network model to predict the location of the device;

[0033] The convolutional neural network model matches the currently scanned Wi-Fi signal with the data in the fingerprint library to obtain the estimated location of the device;

[0034] Step S27: In data fusion and positioning optimization, the Wi-Fi positioning and inertial navigation results are fused through extended Kalman filtering;

[0035] Inertial navigation provides relative position change trends, and Wi-Fi positioning provides absolute position references. The extended Kalman filter uses a state vector to represent the current position and direction of the device, predicts the state through step size and direction, and makes corrections based on Wi-Fi positioning observations.

[0036] Preferably, in step S22, inertial navigation technology is used to estimate the relative change of the device position; first, based on walking step detection, step length estimation and direction calculation; then, each time a step is detected, the system calculates the relative displacement of the device to obtain the current relative position.

[0037] Preferably, in step S23, the step length is calculated based on the Weinberg nonlinear step length model as follows:

[0038]

[0039] Among them, L k is the step length of the kth step; and They represent the maximum and minimum acceleration of the kth pedestrian respectively; α represents the step length estimation arithmetic after least squares fitting.

[0040] Preferably, in step S24, the moving direction calculation formula is as follows:

[0041]

[0042] Among them, φ, θ, ψ represent roll, pitch, and yaw, namely, roll angle, pitch angle, and yaw angle respectively; q w ,q x ,q y ,q z Represents a quaternion provided by an inertial sensor.

[0043] Preferably, in step S25, the position update formula is as follows:

[0044] X k+1 =x k +L k ·sin(ψ k );

[0045] y k+1 =y k +L k ·cos(ψ k );

[0046] Among them, x k ,y k is the device position of the previous step; L k is the step size; ψ k is the direction angle of the current step.

[0047] Preferably, in step S3, a large language model is used to semantically fuse and analyze multimodal data, identify potential risks, and generate a safety assessment report including safety ratings, key findings, and early warning suggestions. The specific process is as follows:

[0048] Step S31: convert the multimodal data into structured information respectively to ensure that it can be efficiently processed by the large language model;

[0049] The positioning data is organized into time series position information and direction angle sequences to form a complete trajectory representation; the event time, type and status information are extracted from the device log; the image data is extracted into a high-dimensional feature vector to represent the changes in visual information; the above data is standardized and unified in format for further analysis;

[0050] Step S32: Encapsulate the sorted multimodal data in JSON format;

[0051] The positioning data, device logs and image features are organized into a JSON data structure in a predefined format. The data of each modality is represented independently while retaining associated information such as timestamps.

[0052] Step S33: design prompt words to clarify task objectives and analysis requirements of the large language model;

[0053] Combined with JSON data, the task background, analysis objectives, and input data description are written into the prompt words so that the large language model can understand the context and output high-quality evaluation results;

[0054] Step S34: calling a large language model to perform semantic analysis and comprehensive processing on the input data;

[0055] The large language model uses built-in multimodal feature fusion and semantic association modules to deeply encode positioning data, device logs, and image features and map them to a shared semantic space. Using the self-attention mechanism, the large language model can dynamically analyze the relationship between multimodal data and automatically discover abnormal patterns and potential risks.

[0056] Step S35: Generate a security assessment report based on the analysis results of the large language model;

[0057] The report is output in natural language, which is concise and intuitive, making it easy for users to quickly understand and apply;

[0058] Step S36: parse the evaluation report returned by the large language model and convert the result into structured content for further processing by the system;

[0059] Based on the key findings and recommendations in the report, trigger corresponding risk warning mechanisms or take protective measures.

[0060] Preferably, in step S35, the report content includes three parts:

[0061] The first is safety rating, which evaluates the overall safety status of industrial assets based on the quantitative analysis of risks using a big language model;

[0062] The second is the key findings, which describe in detail the potential problems or anomalies identified;

[0063] The third is improvement suggestions, which propose specific optimization measures based on key findings.

[0064] Therefore, the present invention adopts the above-mentioned asset positioning monitoring method based on Wi-Fi and inertial navigation fusion and large language model, and the beneficial effects are as follows:

[0065] The present invention innovatively introduces a large language model, and through its multimodal data fusion and semantic analysis capabilities, it achieves an in-depth understanding and dynamic monitoring of the operating status of industrial assets; with the help of this technology, the present invention can efficiently identify potential equipment failures and abnormal conditions, and generate early warning information in a timely manner, effectively breaking through the limitations of traditional single technology and greatly improving the level of intelligence in asset protection.

[0066] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a system block diagram of the asset positioning monitoring method based on Wi-Fi, inertial navigation fusion and large language model of the present invention;

[0068] Figure 2 It is a flow chart of the asset positioning monitoring method based on Wi-Fi, inertial navigation fusion and large language model of the present invention;

[0069] Figure 3 This is a simulation diagram of the fusion positioning trajectory based on Wi-Fi and inertial navigation of the present invention;

[0070] Figure 4 It is a structured safety assessment report obtained by the present invention through data analysis using a large language model;

[0071] Figure 5 This is an example of the design of a large language model prompt word for industrial important asset protection analysis according to the present invention. DETAILED DESCRIPTION

[0072] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0073] like Figure 1 As shown, the asset positioning and monitoring method based on Wi-Fi and inertial navigation fusion and large language model of the present invention combines multimodal data acquisition, Wi-Fi and inertial navigation fusion positioning, multimodal semantic analysis, and anomaly detection and risk assessment modules to achieve accurate positioning and intelligent protection of industrial assets.

[0074] like Figure 2 As shown, the asset positioning monitoring method based on Wi-Fi and inertial navigation fusion and large language model of the present invention includes the following steps:

[0075] Step S1, obtaining multimodal data and building a fingerprint library; wherein the multimodal data includes Wi-Fi signals, inertial navigation data, device logs and image data;

[0076] Step S2: accurately locate industrial assets by using the fusion positioning technology of Wi-Fi and inertial navigation, combining the collected signal strength, inertial navigation path and feature data in the fingerprint library;

[0077] Step S3: Use a large language model to semantically fuse and analyze multimodal data, identify potential risks, and generate a safety assessment report including safety ratings, key findings, and warning recommendations.

[0078] Example

[0079] Step S1, obtaining multimodal data and building a fingerprint library; wherein the multimodal data includes Wi-Fi signals, inertial navigation data, device logs and image data.

[0080] Step S11: Collect data D from different sources using a multimodal data acquisition device t , as shown below:

[0081]

[0082] in, Indicates Wi-Fi signal data, Indicates inertial navigation data, Represents device log and image data; the superscript t represents the sampling time.

[0083] Step S12: Obtain signal strength values ​​D through multiple Wi-Fi access points in the environment WiFi , as shown below:

[0084] D WiFi ={rssi1, rssi2,…, rssi n};

[0085] Among them, D WiFi Represents a set of signal strength values ​​from n access points.

[0086] Step S13: Obtain the inertial navigation data of the device through the inertial sensor, including the linear acceleration D Linear , gravity acceleration D Gravity and the rotation angle D Rotation , as shown below:

[0087] D Linear = {linear x , linear y , linear z};

[0088] D Gravity ={gravity x ,gravity y , gravity z};

[0089] D Rotation ={rotation x , rotation y , rotation z , rotation w};

[0090] The set of the above inertial navigation data is defined as:

[0091] D IN ={D Linear , D Gravity , D Rotation}.

[0092] Step S14: Obtain operation log information from industrial equipment, including timestamp, current status of equipment, operation records, alarm information, and maintenance history, as shown below:

[0093] DLog ={L1, L2, L3, L4, L5}.

[0094] Step S15: Collect real-time image data D through the camera Image Among them, D Image Represents the acquired image data.

[0095] Step S2: accurately locate industrial assets by using the fusion positioning technology of Wi-Fi and inertial navigation, combining the collected signal strength, inertial navigation path and feature data in the fingerprint library.

[0096] Step S21: pre-process the collected multimodal data including Wi-Fi signal strength and inertial navigation data.

[0097] Step S22: Use inertial navigation technology to estimate the relative change of the device position.

[0098] First, based on walking step detection, step length estimation and direction calculation.

[0099] Then, every time a step is detected, the system calculates the relative displacement of the device to derive the current relative position.

[0100] Step S23: During the calculation process, the moving direction and step length of the device are calculated using the output of the inertial sensor, and the position of the device is gradually updated.

[0101] Based on the Weinberg nonlinear step size model, the step size is calculated as follows:

[0102]

[0103] Among them, L k is the step length of the kth step; and They represent the maximum and minimum acceleration of the kth pedestrian respectively; α represents the step length estimation arithmetic after least squares fitting.

[0104] Step S24: Use the rotation vector quaternion integrating the accelerometer, magnetometer and gyroscope data to perform heading calculation.

[0105] The formula for calculating the moving direction is as follows:

[0106]

[0107] Among them, φ, θ, ψ represent roll, pitch, and yaw, namely, roll angle, pitch angle, and yaw angle respectively; q w ,q x ,q y ,q zRepresents a quaternion provided by an inertial sensor.

[0108] Step S25: Based on the previous position (x k ,y k ), step length L k and direction ψ k , calculate the current position of the device (x k+1 ,y k+1 ), so the position update formula is as follows:

[0109] x k+1 =x k +L k ·sin(ψ k );

[0110] y k+1 =y k +L k ·cos(ψ k );

[0111] Among them, x k ,y k is the device position of the previous step; L k is the step size; ψ k is the direction angle of the current step.

[0112] Step S26: combining the real-time Wi-Fi signal strength data and the pre-built fingerprint database, using a convolutional neural network model to predict the location of the device.

[0113] The convolutional neural network model matches the currently scanned Wi-Fi signal with the data in the fingerprint library to obtain the estimated location of the device.

[0114] Step S27: In data fusion and positioning optimization, the Wi-Fi positioning and inertial navigation results are fused through extended Kalman filtering.

[0115] Inertial navigation provides relative position change trends, and Wi-Fi positioning provides absolute position references. The extended Kalman filter uses a state vector to represent the current position and direction of the device, predicts the state through step length and direction, and makes corrections based on Wi-Fi positioning observations.

[0116] Step S3: Use a large language model to semantically fuse and analyze multimodal data, identify potential risks, and generate a safety assessment report including safety ratings, key findings, and warning recommendations.

[0117] Step S31: Convert the multimodal data into structured information respectively to ensure that it can be efficiently processed by the large language model.

[0118] The positioning data is organized into time series position information and direction angle sequence to form a complete trajectory representation; the event time, type and status information are extracted from the device log; the image data is extracted into a high-dimensional feature vector to represent the change of visual information. After the above data is standardized, it is unified in format for further analysis.

[0119] Step S32: Encapsulate the sorted multimodal data in JSON.

[0120] The positioning data, device logs, and image features are organized into a JSON data structure in a predefined format. The data of each modality is represented independently while retaining associated information such as timestamps.

[0121] Step S33: design prompt words to clarify task objectives and analysis requirements of the large language model.

[0122] Combined with JSON data, the task background, analysis objectives, and input data description are written into the prompt words so that the large language model can understand the context and output high-quality evaluation results.

[0123] Step S34: call the large language model to perform semantic analysis and comprehensive processing on the input data.

[0124] The large language model deeply encodes the location data, device logs and image features through the built-in multimodal feature fusion and semantic association module, and maps them to a shared semantic space. Using the self-attention mechanism, the large language model can dynamically analyze the relationship between multimodal data and automatically discover abnormal patterns and potential risks.

[0125] Step S35: Generate a security assessment report based on the analysis results of the large language model.

[0126] The report consists of three parts: one is the safety rating, which evaluates the overall safety status of industrial assets based on the quantitative analysis of risks using a large language model; the second is the key findings, which describes in detail the potential problems or anomalies identified; and the third is the improvement suggestions, which propose specific optimization measures for the key findings.

[0127] The report is output in natural language, which is concise and intuitive, making it easy for users to quickly understand and apply it.

[0128] Step S36: parse the evaluation report returned by the large language model and convert the results into structured content for further processing by the system.

[0129] Based on the key findings and recommendations in the report, trigger corresponding risk warning mechanisms or take protective measures.

[0130] In order to verify the authenticity of the present invention, an experimental simulation was conducted, and the experimental data was based on the real collected Wi-Fi fingerprint and inertial navigation database as well as the self-built historical log and collected image database.

[0131] like Figure 3 As shown in the figure, using extended Kalman filtering to fuse Wi-Fi and inertial navigation data can greatly improve positioning accuracy and meet the precise positioning requirements for industrial asset protection.

[0132] like Figure 4 As shown in the figure, a structured safety assessment report can be obtained by analyzing the data using a large language model, which shows that the large language model can efficiently integrate and interpret complex industrial data and provide strong support for safety analysis.

[0133] like Figure 5 As shown in the figure, through clear task description and data requirements, clear guidance is provided for the semantic fusion and comprehensive analysis of multimodal data.

[0134] Therefore, the present invention adopts the above-mentioned asset positioning monitoring method based on Wi-Fi and inertial navigation fusion and large language model, which can efficiently integrate and analyze multimodal data and quickly identify potential safety hazards in industrial assets. By introducing the semantic analysis and anomaly detection capabilities of the large language model, this method significantly improves the monitoring and risk identification capabilities of the operating status of industrial assets, and provides more comprehensive and intelligent support for the safety analysis of important industrial assets.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. An asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and large language model, characterized in that: The following steps are involved: Step S1, obtaining multimodal data and building a fingerprint library; wherein the multimodal data includes Wi-Fi signals, inertial navigation data, device logs and image data; Step S2: accurately locate industrial assets by using the fusion positioning technology of Wi-Fi and inertial navigation, combining the collected signal strength, inertial navigation path and feature data in the fingerprint library; Step S3: Use a large language model to semantically fuse and analyze multimodal data, identify potential risks, and generate a safety assessment report including safety ratings, key findings, and warning recommendations.

2. The asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and large language model according to claim 1 is characterized in that: In step S1, multimodal data is obtained to build a fingerprint library; wherein the multimodal data includes Wi-Fi signals, inertial navigation data, device logs and image data. The specific process is as follows: Step S11: Collect data D from different sources using a multimodal data acquisition device t , as shown below: in, Indicates Wi-Fi signal data, Indicates inertial navigation data, Indicates device log and image data; the superscript t indicates the sampling time; Step S12: Obtain signal strength values ​​D through multiple Wi-Fi access points in the environment WiFi , as shown below: D WiFi ={rssi1,rssi2,…,rssi n }; Among them, D WiFi represents a set of signal strength values ​​from n access points; Step S13: Obtain the inertial navigation data of the device through the inertial sensor, including the linear acceleration D Linear , gravity acceleration D Gravity and the rotation angle D Rotation , as shown below: D Linear ={linear x ,linear y ,linear z }; D Gravity ={gravity x ,gravity y ,gravity z }; D Rotation ={rotation x ,rotation y ,rotation z ,rotation w }; The set of the above inertial navigation data is defined as: D IN ={D Linear ,D Gravity ,D Rotation }; Step S14: Obtain operation log information from industrial equipment, including timestamp, current status of equipment, operation records, alarm information, and maintenance history, as shown below: <h2 style=";text-align:left;direction:ltr">D<h2 style=";text-align:left;direction:ltr"> Log <h2 style=";text-align:left;direction:ltr"> (L1,L2,L3,L4,L5) Step S15: Collect real-time image data D through the camera Image Among them, D Image Represents the acquired image data.

3. The asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and large language model according to claim 1 is characterized in that: In step S2, the industrial assets are accurately located by using the fusion positioning technology of Wi-Fi and inertial navigation, combining the collected signal strength, inertial navigation path and feature data in the fingerprint library. The specific process is as follows: Step S21, preprocessing the collected multimodal data including Wi-Fi signal strength and inertial navigation data; Step S22: using inertial navigation technology to estimate the relative change of the device position; Step S23: During the calculation process, the moving direction and step length of the device are calculated using the output of the inertial sensor, and the position of the device is gradually updated; Step S24, using the rotating vector quaternion of the integrated accelerometer, magnetometer and gyroscope data to perform heading calculation; Step S25: Based on the previous position (x k ,y k ), step length L k and direction ψ k , calculate the current position of the device (x k+1 ,y k+1 ); Step S26: combining the real-time Wi-Fi signal strength data and the pre-built fingerprint database, using a convolutional neural network model to predict the location of the device; The convolutional neural network model matches the currently scanned Wi-Fi signal with the data in the fingerprint library to obtain the estimated location of the device; Step S27: In data fusion and positioning optimization, the Wi-Fi positioning and inertial navigation results are fused through extended Kalman filtering; Inertial navigation provides relative position change trends, and Wi-Fi positioning provides absolute position references. The extended Kalman filter uses a state vector to represent the current position and direction of the device, predicts the state through step size and direction, and makes corrections based on Wi-Fi positioning observations.

4. The asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and large language model according to claim 3 is characterized by: In step S22, inertial navigation technology is used to estimate the relative change of the device position; first, based on walking step detection, step length estimation and direction calculation; then, each time a step is detected, the system calculates the relative displacement of the device to obtain the current relative position.

5. The asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and large language model according to claim 3 is characterized in that: In step S23, the step size is calculated based on the Weinberg nonlinear step size model as follows: Among them, L k is the step length of the kth step; and They represent the maximum and minimum acceleration of the kth pedestrian respectively; α represents the step length estimation arithmetic after least squares fitting.

6. The asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and large language model according to claim 3 is characterized in that: In step S24, the moving direction calculation formula is as follows: Among them, φ, θ, ψ represent roll, pitch, and yaw, namely, roll angle, pitch angle, and yaw angle respectively; q w ,q x ,q y ,q z Represents a quaternion provided by an inertial sensor.

7. The asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and large language model according to claim 3 is characterized in that: In step S25, the position update formula is as follows: x k+1 =x k +L k ·sin(ψ k ); and k+1 =and k +L k ·cos(ψ k ); Among them, x k ,y k is the device position of the previous step; L k is the step size; ψ k is the direction angle of the current step.

8. The asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and large language model according to claim 1 is characterized in that: In step S3, a large language model is used to semantically fuse and analyze multimodal data, identify potential risks, and generate a safety assessment report, including safety ratings, key findings, and warning recommendations. The specific process is as follows: Step S31: convert the multimodal data into structured information respectively to ensure that it can be efficiently processed by the large language model; The positioning data is organized into time series position information and direction angle sequences to form a complete trajectory representation; the event time, type and status information are extracted from the device log; the image data is extracted into a high-dimensional feature vector to represent the changes in visual information; the above data is standardized and unified in format for further analysis; Step S32: Encapsulate the sorted multimodal data in JSON format; The positioning data, device logs and image features are organized into a JSON data structure in a predefined format. The data of each modality is represented independently while retaining associated information such as timestamps. Step S33: design prompt words to clarify task objectives and analysis requirements of the large language model; Combined with JSON data, the task background, analysis objectives, and input data description are written into the prompt words so that the large language model can understand the context and output high-quality evaluation results; Step S34: calling a large language model to perform semantic analysis and comprehensive processing on the input data; The large language model uses built-in multimodal feature fusion and semantic association modules to deeply encode positioning data, device logs, and image features and map them to a shared semantic space. Using the self-attention mechanism, the large language model can dynamically analyze the relationship between multimodal data and automatically discover abnormal patterns and potential risks. Step S35: Generate a security assessment report based on the analysis results of the large language model; The report is output in natural language, which is concise and intuitive, making it easy for users to quickly understand and apply; Step S36: parse the evaluation report returned by the large language model and convert the result into structured content for further processing by the system; Based on the key findings and recommendations in the report, trigger corresponding risk warning mechanisms or take protective measures.

9. The asset positioning and monitoring method based on Wi-Fi, inertial navigation fusion and large language model according to claim 8 is characterized in that: In step S35, the report content includes three parts: The first is safety rating, which evaluates the overall safety status of industrial assets based on the quantitative analysis of risks using a big language model; The second is the key findings, which describe in detail the potential problems or anomalies identified; The third is improvement suggestions, which propose specific optimization measures based on key findings.

Images