Explosion-proof intelligent terminal positioning method and device based on high-end displacement sensor

Abstract

The present invention relates to the field of sensor positioning technology, and discloses an explosion-proof intelligent terminal positioning method and device based on a high-end displacement sensor, the method comprising: collecting data from the explosion-proof intelligent terminal to obtain an original sensor data set; filtering and calibration processing to obtain inertial measurement unit data, geomagnetic field sensor data and wireless signal receiver data; performing double integral operations to obtain initial relative displacement data; measuring absolute displacement measurement data; performing position analysis through a multi-level Kalman filter model to obtain terminal position data; performing three-dimensional spatial positioning through a multi-source data fusion model to obtain a three-dimensional spatial positioning result of the explosion-proof intelligent terminal. The present invention can effectively fuse multi-source sensor data, overcome the limitations of each single positioning technology, and achieve real-time and stable positioning under limited computing resources. At the same time, explosion-proof safety requirements are also taken into consideration to ensure reliable application in dangerous environments.

Description

Technical Field

[0001] The present invention relates to the field of sensor positioning technology, and in particular to an explosion-proof intelligent terminal positioning method and device based on a high-end displacement sensor. Background Art

[0002] In industrial production and hazardous environments, accurate positioning of explosion-proof smart terminals is crucial for safety management and efficient operation. Traditional positioning methods mainly rely on technologies such as GPS, Wi-Fi or Bluetooth, but these methods often have difficulty ensuring positioning accuracy indoors or in signal-restricted environments. At the same time, the presence of a large number of metal devices and electromagnetic interference in complex industrial environments further reduces the reliability of wireless signals.

[0003] Although the inertial navigation system can work in a signal-free environment, the accumulated sensor errors will cause positioning drift after long-term use. Although geomagnetic field positioning is not affected by signal shielding, it is easily interfered by surrounding metal objects. Existing fusion positioning algorithms are usually unable to effectively process multi-source heterogeneous data, resulting in insufficient positioning accuracy and reliability. In addition, in explosion-proof environments, the power consumption and heat generation of equipment also limit the application of complex algorithms. Summary of the invention

[0004] The present invention provides an explosion-proof intelligent terminal positioning method and device based on high-end displacement sensors. The present invention can effectively integrate multi-source sensor data, overcome the limitations of each single positioning technology, and achieve real-time and stable positioning under limited computing resources. At the same time, explosion-proof safety requirements are also taken into account to ensure reliable application in dangerous environments.

[0005] In a first aspect, the present invention provides an explosion-proof intelligent terminal positioning method based on a high-end displacement sensor, the explosion-proof intelligent terminal positioning method based on a high-end displacement sensor comprising:

[0006] Collect data from the inertial measurement unit, geomagnetic field sensor and wireless signal receiver in the explosion-proof intelligent terminal to obtain the original sensor data set;

[0007] Filtering and calibrating the raw sensor data set to obtain inertial measurement unit data, geomagnetic field sensor data, and wireless signal receiver data;

[0008] Performing a double integral operation based on the inertial measurement unit data to obtain initial relative displacement data of the explosion-proof intelligent terminal;

[0009] The position of the explosion-proof intelligent terminal relative to a fixed reference point is measured by a displacement sensor to obtain absolute displacement measurement data;

[0010] Inputting the initial relative displacement data and the absolute displacement measurement data into a multi-level Kalman filter model for position analysis to obtain terminal position data;

[0011] The terminal location data, the geomagnetic field sensor data and the wireless signal receiver data are input into a multi-source data fusion model for three-dimensional spatial positioning to obtain a three-dimensional spatial positioning result of the explosion-proof intelligent terminal.

[0012] In a second aspect, the present invention provides an explosion-proof intelligent terminal positioning device based on a high-end displacement sensor, the explosion-proof intelligent terminal positioning device based on a high-end displacement sensor comprises:

[0013] The acquisition module is used to collect data from the inertial measurement unit, geomagnetic field sensor and wireless signal receiver in the explosion-proof intelligent terminal to obtain the original sensor data set;

[0014] A calibration module, used for filtering and calibrating the raw sensor data set to obtain inertial measurement unit data, geomagnetic field sensor data and wireless signal receiver data;

[0015] A calculation module, used for performing a double integral operation based on the inertial measurement unit data to obtain the initial relative displacement data of the explosion-proof intelligent terminal;

[0016] A measuring module, used to measure the position of the explosion-proof intelligent terminal relative to a fixed reference point through a displacement sensor to obtain absolute displacement measurement data;

[0017] An analysis module, used for inputting the initial relative displacement data and the absolute displacement measurement data into a multi-level Kalman filter model for position analysis to obtain terminal position data;

[0018] The positioning module is used to input the terminal location data, the geomagnetic field sensor data and the wireless signal receiver data into a multi-source data fusion model for three-dimensional spatial positioning to obtain a three-dimensional spatial positioning result of the explosion-proof intelligent terminal.

[0019] The third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned explosion-proof intelligent terminal positioning method based on high-end displacement sensors.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned explosion-proof intelligent terminal positioning method based on a high-end displacement sensor.

[0021] In the technical solution provided by the present invention, the data of the inertial measurement unit, the geomagnetic field sensor, the wireless signal receiver and the high-end displacement sensor are combined, and a comprehensive analysis is performed through a multi-source data fusion model, which effectively improves the accuracy and robustness of positioning. The introduction of a high-end displacement sensor for absolute displacement measurement provides an accurate reference point for the inertial navigation system, effectively suppresses the cumulative error, and significantly improves the accuracy of long-term positioning. The use of technologies such as dynamic window division and adaptive zero-speed update can automatically adjust the data processing strategy according to the motion state of the terminal, thereby improving the adaptability and accuracy of the algorithm. A multi-level Kalman filter model including position, velocity and acceleration estimation layers is designed, and a more stable and accurate state estimation is achieved through inter-layer information transmission and correction. The introduction of convolutional neural networks and attention mechanisms in the multi-source data fusion model can automatically learn and extract the key features of multi-source data, thereby improving the fusion effect. The spatial constraint optimization of the initial three-dimensional position coordinates is performed by the conditional random field, and the rationality of the positioning results is further improved by combining the prior knowledge of the environment. The neural network structure is optimized by batch normalization, dropout and other technologies, which reduces the computational complexity while ensuring the positioning accuracy, and is suitable for running on resource-constrained explosion-proof intelligent terminals. The present invention takes into account the special requirements of explosion-proof environments, avoids the problems of high power consumption and heating, and ensures safe application in hazardous environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 Schematic diagram of the steps of the explosion-proof intelligent terminal positioning method based on the high-end displacement sensor in an embodiment of the present invention;

[0024] Figure 2 Schematic diagram of the structure of an explosion-proof intelligent terminal positioning device based on a high-end displacement sensor in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The embodiment of the present invention provides an explosion-proof intelligent terminal positioning method and device based on a high-end displacement sensor. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 In an embodiment of the present invention, an explosion-proof intelligent terminal positioning method based on a high-end displacement sensor includes:

[0027] Step S1, collecting data from an inertial measurement unit, a geomagnetic field sensor, and a wireless signal receiver in an explosion-proof intelligent terminal to obtain an original sensor data set;

[0028] It is understandable that the execution subject of the present invention may be an explosion-proof intelligent terminal positioning device based on a high-end displacement sensor, or a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0029] Specifically, the accelerometer and gyroscope in the inertial measurement unit are sampled at high frequency to obtain raw acceleration data and raw angular velocity data. Since the inertial sensor will accumulate errors when used for a long time, zero speed detection is performed based on the raw acceleration data and raw angular velocity data. Through zero speed detection, the moment when the terminal is in a stationary state can be identified to obtain a stationary state mark sequence. The raw acceleration data and raw angular velocity data are segmented according to the stationary state mark sequence. The raw sensor data is dynamically windowed, and the data is divided into more timely small segments by combining the detected stationary state sequence, thereby improving the accuracy of subsequent data processing. At the same time, the geomagnetic field sensor in the explosion-proof intelligent terminal is sampled. The geomagnetic field sensor detects the geomagnetic field strength around the explosion-proof intelligent terminal and generates segmented geomagnetic field strength data corresponding to the dynamic window by combining the results of dynamic window division. At the same time, the signal strength of the surrounding WiFi access points and UWB base stations is collected based on the wireless signal receiver in the explosion-proof intelligent terminal to obtain a wireless signal strength data set. Since WiFi and UWB signals can provide additional spatial information, three-dimensional spatial positioning provides more data dimensions and accuracy. The raw acceleration data and raw angular velocity data obtained from the accelerometer and gyroscope, the segmented geomagnetic field strength data, and the wireless signal strength data set are time synchronized and aligned to ensure that all data are corresponding in the time dimension, that is, to ensure that each data point can reflect the motion state and spatial information of the terminal within the same time period, thereby generating a complete raw sensor data set.

[0030] Step S2, filtering and calibrating the original sensor data set to obtain inertial measurement unit data, geomagnetic field sensor data and wireless signal receiver data;

[0031] Specifically, the original acceleration data is subjected to wavelet threshold denoising to effectively filter out the high-frequency noise in the acceleration data and obtain the denoised acceleration data. Based on the denoised acceleration data, the gravity component is calculated, and the linear acceleration data is obtained by removing the gravity component. The original angular velocity data is filtered to eliminate the noise and error in the angular velocity data and obtain the filtered angular velocity data. Based on the linear acceleration data and the filtered angular velocity data, attitude solution is performed, and the attitude quaternion of the terminal is obtained by using a common attitude estimation algorithm (such as the quaternion method). At the same time, the ellipsoid fitting algorithm is applied to the original segmented geomagnetic field intensity data to calibrate the soft iron and hard iron effects. Through the soft iron and hard iron calibration, the magnetic field distortion introduced by the sensor and the environment is effectively compensated to obtain the calibrated geomagnetic field data. Based on the calibrated geomagnetic field data and the attitude quaternion calculated previously, the coordinate system conversion algorithm is used to convert the geomagnetic data from the sensor coordinate system to the geographic coordinate system, and the geomagnetic north vector is calculated to provide an important direction reference for subsequent positioning. In terms of processing wireless signal data, outliers are removed from the wireless signal strength data set to filter out extreme data points caused by environmental interference or equipment abnormalities. The signal strength data set after outliers are removed is more realistic and reliable. Based on the distance attenuation model, the wireless signal strength data is converted into distance data. Through this model, the distance between each WiFi access point or UWB base station and the explosion-proof intelligent terminal is estimated. Data division and data alignment are performed on linear acceleration data, filtered angular velocity data, geomagnetic north vector, and wireless signal distance data to ensure that data from different sensors can be accurately aligned in the same time period, avoid data inconsistency caused by time difference, and obtain stable and high-precision inertial measurement unit data, geomagnetic field sensor data, and wireless signal receiver data.

[0032] Step S3: performing double integration operation based on the inertial measurement unit data to obtain initial relative displacement data of the explosion-proof intelligent terminal;

[0033] Specifically, the linear acceleration data in the inertial measurement unit data is converted into a coordinate system, and the linear acceleration data is converted into a geographic coordinate system through the attitude quaternion to ensure that the data can reflect the motion state of the terminal in the actual space. The acceleration data in the geographic coordinate system is subjected to dynamic threshold analysis to identify the user's gait cycle, and each gait cycle is divided by detecting the characteristic changes of the steps in the acceleration data to obtain the single-step acceleration data. The single-step acceleration data is adaptively updated to zero speed. By detecting the user's gait state, the speed is updated to zero at the end position of each step to eliminate the error introduced by sensor drift or noise, and the corrected acceleration data is obtained. Based on the corrected acceleration data, the first integral operation is performed to obtain the speed data. The speed data is often affected by the cumulative error during the integration process, and the speed data is corrected for drift error. Drift error correction is a compensation method. By detecting and correcting the drift in the speed data, the corrected speed data obtained in the end can better reflect the actual motion situation. A second integral operation is performed based on the corrected speed data to obtain displacement data. By quadratically integrating the velocity data, the displacement change of the terminal in space can be calculated. The displacement data is obtained by integrating the velocity data and represents the relative displacement of the terminal in the geographic coordinate system. The displacement data is accumulated to ensure that the displacement of each step can be gradually superimposed to form a complete moving trajectory. At the same time, combined with the attitude quaternion in the inertial measurement unit, all displacement data are unified to the geographic coordinate system through coordinate transformation, and finally the initial relative displacement data of the explosion-proof intelligent terminal is obtained.

[0034] Step S4, measuring the position of the explosion-proof intelligent terminal relative to the fixed reference point by a displacement sensor to obtain absolute displacement measurement data;

[0035] Specifically, the displacement sensor in the explosion-proof intelligent terminal is initialized and calibrated to obtain the initial parameters of the sensor. The initialization calibration process includes calibrating the sensor in a stationary or known environment to determine the zero offset, gain and linear response characteristics of the sensor. Based on the initial parameters of the sensor, the raw data collected by the displacement sensor is linearized to eliminate the measurement error caused by the nonlinear response of the sensor, and the raw displacement data is converted into data that is linearly related to the actual displacement to obtain linearized displacement data. The linearized displacement data is adaptively median filtered to effectively remove instantaneous spike noise or other extreme values ​​in the data to obtain smooth and stable filtered displacement data. Fourier transform is performed based on the filtered displacement data to convert the time domain signal into a frequency domain signal to obtain the frequency domain characteristics of the displacement data. The frequency domain characteristics can reveal hidden information such as periodic components and high-frequency noise in the data. By performing peak detection and phase analysis on the frequency domain feature data, the key motion features in the displacement data are identified and a more accurate displacement is calculated. Peak detection is used to identify the maximum change amplitude in the displacement signal, reflecting the distance and speed changes of the terminal movement, while phase analysis is used to determine the spatial position and movement direction of the terminal relative to a fixed reference point. Based on the precise displacement and combined with the preset fixed reference point, coordinate transformation is performed to finally obtain relative position data, and the displacement measured by the sensor is converted into a relative position in the actual space, so that the displacement data is associated with the actual physical reference point. The relative position data is processed using noise suppression technology to eliminate high-frequency noise and low-frequency interference, and obtain smoother filtered relative position data. Based on the filtered relative position data and combined with the posture information of the explosion-proof intelligent terminal, the coordinate system alignment operation is performed. The posture information can help correct the data deviation of the displacement sensor in different spatial orientations. Through coordinate system alignment, the relative position data is unified into the global coordinate system, and finally accurate absolute displacement measurement data is obtained.

[0036] Step S5, inputting the initial relative displacement data and the absolute displacement measurement data into a multi-level Kalman filter model for position analysis to obtain terminal position data;

[0037] Specifically, the initial relative displacement data and the absolute displacement measurement data are preprocessed. The preprocessing process includes two key steps: coordinate system conversion and time synchronization. The coordinate system conversion is to ensure that data from different sources are operated under a unified reference system to avoid errors caused by coordinate system differences. Time synchronization is to ensure that the timestamps of different data are consistent so that the relative displacement data and the absolute displacement data can be aligned at the same time point. Through the preprocessing operation, the preprocessed relative displacement data and absolute displacement data are obtained. The preprocessed relative displacement data and absolute displacement data are input into a multi-level Kalman filter model, which is divided into three levels: position estimation layer, velocity estimation layer and acceleration estimation layer. In the position estimation layer, the preprocessed relative displacement data is used as a state vector input, and the preprocessed absolute displacement data is used as an observation vector input. Through the first state prediction module of the position estimation layer, the state prediction of these data is performed to obtain preliminary position information, that is, the position prediction value. In the first measurement update module of the position estimation layer, the first Kalman gain is calculated using the position prediction value and the observed position information. Kalman gain is a weighting coefficient used to find the optimal solution between state prediction and observation results. In the first state fusion module, the position prediction value is corrected based on the calculated Kalman gain to obtain accurate position information, that is, the position state estimation value. In the velocity estimation layer, the first-order difference operation is performed based on the preprocessed relative displacement data to calculate the velocity data. The velocity data is input as the state vector to the state prediction module of the velocity estimation layer, and the velocity prediction value is obtained through state prediction. Similar to the position estimation layer, in the second measurement update module of the velocity estimation layer, the second Kalman gain is calculated based on the velocity prediction value and the observed velocity value. In the second state fusion module of the velocity estimation layer, the velocity prediction value is corrected using the gain to obtain a more accurate velocity state estimation value. In the acceleration estimation layer, the velocity data is first-order differentiated to calculate the acceleration data. The acceleration data is input as the state vector to the state prediction module of the acceleration estimation layer, and the acceleration prediction value is obtained through prediction. In the third measurement update module of the acceleration estimation layer, the third Kalman gain is calculated based on the acceleration prediction value and the observed acceleration value. Through this gain, the acceleration prediction value is corrected in the third state fusion module of the acceleration estimation layer to obtain the final acceleration state estimation value. Weighted fusion is performed based on the position state estimation value, velocity state estimation value and acceleration state estimation value. By integrating the data results of each level, more accurate and reliable terminal position data is obtained. Through the processing of multi-level Kalman filtering, the system can effectively integrate multi-source data, not only to correct the errors caused by inertial measurement, but also to make full use of absolute displacement data, and finally realize the precise positioning of explosion-proof intelligent terminals in complex environments.

[0038] Among them, in the first state prediction module of the position estimation layer, the state vector of the preprocessed relative displacement data is initialized to obtain the initial state vector and the initial covariance matrix. The initial state vector reflects the relative position of the explosion-proof intelligent terminal at the current time step, while the initial covariance matrix represents the uncertainty of the state estimation. Based on the initial state vector and the initial covariance matrix, the state transfer matrix is ​​used to predict the state and generate the prior state vector. The state transfer matrix is ​​one of the cores of the Kalman filter, which updates the position information by predicting the state change at the next moment. At the same time, based on the process noise covariance matrix, the covariance prediction of the prior state vector is performed to obtain the prior covariance matrix, evaluate the uncertainty in the prediction process, and ensure that the noise and error can be reasonably controlled during the state estimation process. After completing the state prediction, the observation vector is constructed based on the preprocessed absolute displacement data, and the prior state vector is mapped to the observation space through the observation matrix to obtain the predicted observation vector. The role of the observation matrix is ​​to link the state variables with the observation data and convert the state vector into a quantity that can be directly compared with the actual observation data. The residual of the predicted observation vector and the actual observation vector is calculated to obtain the observation residual, which reflects the difference between the system prediction value and the actual observation value, indicating the deviation of the model in the current position estimation. In the first measurement update module of the position estimation layer, the first Kalman gain is calculated based on the prior covariance matrix, the observation matrix and the observation noise covariance matrix. The Kalman gain plays a role in balancing the predicted value and the observed value in the filtering process, and determines the proportion of the fusion of the predicted value and the observed value. The higher the calculated Kalman gain, the greater the weight of the observation data in the state estimation; on the contrary, if the gain is low, the system relies more on the state prediction value. In the first state fusion module of the position estimation layer, the first Kalman gain and the observation residual are used to correct the prior state vector to obtain the posterior state vector. Through this correction, the predicted value and the observed value can be effectively fused to improve the accuracy of the position estimation. At the same time, based on the first Kalman gain, the observation matrix and the prior covariance matrix, the covariance matrix is ​​updated to obtain the posterior covariance matrix. The posterior covariance matrix characterizes the uncertainty of the updated state estimation and reflects the accuracy of the corrected state estimation. The posterior state vector is output as the position state estimate value, representing the accurate position estimation result after being processed by the multi-level Kalman filter model. Through the process of state prediction, measurement update and state fusion, the system can gradually reduce the noise and error in the position estimation and achieve high-precision positioning of the explosion-proof intelligent terminal.

[0039] Step S6: Input the terminal location data, geomagnetic field sensor data and wireless signal receiver data into a multi-source data fusion model for three-dimensional spatial positioning to obtain a three-dimensional spatial positioning result of the explosion-proof intelligent terminal.

[0040] Specifically, the terminal location data, geomagnetic field sensor data and wireless signal receiver data are time synchronized and aligned. Time synchronization ensures that the data is processed under the same time reference to avoid errors caused by time deviation. Data alignment is to ensure that the data from multiple sensors can be matched in space and effectively combined to form aligned multi-source data. The aligned multi-source data is input into the feature extraction layer of the multi-source data fusion model, in which a convolutional neural network is used to extract features from the input data. The convolutional neural network can automatically extract useful spatial and temporal features from complex multi-source data and generate a multi-source feature map. This multi-source feature map reflects the multi-dimensional data characteristics such as location information, direction information and signal strength provided by different sensors. After obtaining the feature map, the self-attention mechanism is applied to analyze and calculate the correlation weights between the features. Through the self-attention mechanism, it is identified which features are more important in the positioning task, and corresponding weights are assigned to each feature to obtain a weighted feature map. The weighted feature map is input into the fusion layer of the multi-source data fusion model. The fusion layer interacts and fuses features from different sources through a cross-attention mechanism. This process allows the data from different sensors to refer to and adjust each other at the feature level, so that the system can make comprehensive use of multi-source information to generate fusion features. After the fusion features are generated, residual connections and layer normalization are performed on them. Residual connections ensure that information can be transmitted losslessly in the model and reduce the gradient vanishing problem in deep networks, while layer normalization helps stabilize the training process, improve the accuracy and consistency of fusion features, and obtain optimized fusion features. The optimized fusion features are input into the three-dimensional space mapping layer of the multi-source data fusion model. In this layer, the optimized fusion features are mapped to the three-dimensional space coordinate system through a fully connected neural network to generate initial three-dimensional position coordinates. The role of the fully connected neural network is to convert complex features into clear spatial position information to ensure that the fused data can reflect the initial position of the explosion-proof intelligent terminal in three-dimensional space. The initial three-dimensional position coordinates are optimized by conditional random field (CRF). The spatial constraint optimization of CRF can ensure that the three-dimensional position coordinates are consistent in space, avoid physically unreasonable jumps or errors, and obtain optimized three-dimensional position coordinates. The optimized three-dimensional position coordinates are input into the Kalman smoother to smooth the position data of multiple time steps to ensure that the terminal's motion trajectory in the time dimension is more continuous and stable, and to generate accurate three-dimensional spatial positioning results for the explosion-proof intelligent terminal.

[0041] Among them, the optimized fusion features are input to the first fully connected sublayer of the three-dimensional space mapping layer. This sublayer performs nonlinear transformation on the input features through the ReLU activation function. The ReLU activation function helps to introduce nonlinearity, enhance the expression ability of the model, and generate the output of the first hidden layer. In order to ensure that the model remains stable and accelerates convergence during training, the output of the first hidden layer is batch normalized, the output data is standardized, the distribution differences in different feature dimensions are eliminated, and the standardized output of the first hidden layer is obtained. The standardized output of the first hidden layer is input to the second fully connected sublayer of the three-dimensional space mapping layer. In this sublayer, the Leaky ReLU activation function is used to extract features from the data. Leaky ReLU is an improved version of ReLU that allows small negative gradients to pass through, thereby solving the "neuron death" problem that may occur when traditional ReLU processes negative values. In this way, the model can better extract complex features. After obtaining the output of the second hidden layer, it is subjected to dropout regularization. Dropout regularization prevents overfitting of the model by randomly discarding some neurons, enhances the generalization ability of the model, and obtains the regularized output of the second hidden layer. The regularized output of the second hidden layer is input to the third fully connected sublayer of the three-dimensional space mapping layer. In this layer, the Tanh activation function is used for feature mapping. The Tanh activation function compresses the data to the interval [-1, 1], which helps to process spatial data with strong continuity while maintaining the smoothness and derivability of the output. Through the activation of Tanh, the output of the third hidden layer is generated as the result of feature mapping. In order to enhance the expressive power of the model, the output of the third hidden layer is residually connected. The output of the third hidden layer is added to the previously optimized fusion features to form a residual enhanced feature representation. The residual connection helps to avoid the gradient vanishing problem and allows information to jump directly in the network, so that the network can more effectively retain the original input features and improve the model's fitting ability and training efficiency. The residual enhanced feature representation is input to the output layer of the three-dimensional space mapping layer. The output layer maps the features from the high-dimensional space to the three-dimensional space coordinate system through linear transformation. The role of the linear transformation is to convert the output of the network into a clear three-dimensional coordinate value, which represents the initial three-dimensional position coordinates of the explosion-proof intelligent terminal in the actual space.

[0042] In the embodiment of the present invention, the data of the inertial measurement unit, the geomagnetic field sensor, the wireless signal receiver and the high-end displacement sensor are combined, and a comprehensive analysis is performed through a multi-source data fusion model, which effectively improves the accuracy and robustness of positioning. The introduction of a high-end displacement sensor for absolute displacement measurement provides an accurate reference point for the inertial navigation system, effectively suppresses the cumulative error, and significantly improves the accuracy of long-term positioning. The use of technologies such as dynamic window division and adaptive zero-speed update can automatically adjust the data processing strategy according to the motion state of the terminal, thereby improving the adaptability and accuracy of the algorithm. A multi-level Kalman filter model including position, velocity and acceleration estimation layers is designed, and a more stable and accurate state estimation is achieved through inter-layer information transmission and correction. The introduction of convolutional neural networks and attention mechanisms in the multi-source data fusion model can automatically learn and extract the key features of multi-source data, thereby improving the fusion effect. The spatial constraint optimization of the initial three-dimensional position coordinates is performed through conditional random fields, and the rationality of the positioning results is further improved by combining environmental prior knowledge. The neural network structure is optimized by batch normalization, dropout and other technologies, which reduces the computational complexity while ensuring the positioning accuracy, and is suitable for running on resource-constrained explosion-proof intelligent terminals. The present invention takes into account the special requirements of explosion-proof environments, avoids the problems of high power consumption and heating, and ensures safe application in hazardous environments.

[0043] In a specific embodiment, the process of executing step S1 may specifically include the following steps:

[0044] Perform high-frequency sampling on the accelerometer and gyroscope in the inertial measurement unit in the explosion-proof intelligent terminal to obtain raw acceleration data and raw angular velocity data;

[0045] Perform zero velocity detection based on the original acceleration data and the original angular velocity data to obtain a stationary state mark sequence;

[0046] The original acceleration data and the original angular velocity data are segmented according to the stationary state marking sequence to obtain a dynamic window division result;

[0047] The geomagnetic field sensor in the explosion-proof intelligent terminal is sampled and combined with the dynamic window division result to obtain the segmented geomagnetic field intensity data;

[0048] Based on the wireless signal receiver in the explosion-proof intelligent terminal, the signal strength of the surrounding WiFi access points and UWB base stations is collected to obtain a wireless signal strength data set;

[0049] The original acceleration data, the original angular velocity data, the segmented geomagnetic field strength data and the wireless signal strength data set are time synchronized and data aligned to obtain the original sensor data set.

[0050] Specifically, the accelerometer and gyroscope are initialized and set to ensure that they can continuously collect data in extremely short time intervals. The purpose of high-frequency sampling is to capture every subtle change in acceleration and angular velocity of the explosion-proof smart terminal during movement. The accelerometer can measure the linear acceleration of the terminal in three-dimensional space, while the gyroscope can capture the rotational angular velocity information. Together, the two constitute the core data source of the inertial measurement unit. When the accelerometer performs high-frequency sampling, raw acceleration data is generated to record the acceleration changes of the terminal on the x, y, and z axes. Similarly, the gyroscope also generates raw angular velocity data to reflect the changes in the rotational speed of the terminal in each axis. These raw data are usually presented as time series, in which each time point contains acceleration values ​​in three directions and angular velocity values ​​in three directions. Zero speed detection is performed based on the raw data to identify the time point when the terminal is stationary during movement. Zero speed detection is usually based on the amplitude changes of acceleration and angular velocity. When the acceleration value is close to the gravitational acceleration g (about 9.8m / s²) and the angular velocity is close to zero, the system can determine that the terminal is stationary. Use the following formula to calculate the modulus of acceleration and angular velocity at a certain point in time:

[0051] ;

[0052] in, Indicates time point The acceleration modulus, , and are the acceleration components of the x, y, and z axes respectively. Close to 9.8m / s² and angular velocity When it is close to zero, the system can determine that the terminal is stationary at this time. By analyzing the changes in acceleration and angular velocity over a period of time, a stationary state mark sequence is generated to indicate whether the terminal is in a stationary state at each time point. Based on the stationary state mark sequence, the original acceleration data and the original angular velocity data are segmented to separate the motion and stationary states to form different dynamic windows. Each dynamic window represents a continuous motion or stationary state. Through dynamic window division, data in different states are processed in a targeted manner. For example, when the terminal is stationary, it is regarded as a zero displacement interval, while when it is moving, it is necessary to further analyze the changes in acceleration and angular velocity to calculate the displacement and rotation angle of the terminal. At the same time, the geomagnetic field sensor in the explosion-proof smart terminal regularly samples the surrounding geomagnetic field intensity. The geomagnetic field sensor can measure the components of the earth's magnetic field in the x, y, and z directions to generate a three-dimensional magnetic field intensity vector. When combined with the dynamic window division results, the sampling frequency or segmentation method of the geomagnetic field data is adjusted according to the changes in the acceleration and angular velocity data in each window to obtain the segmented geomagnetic field intensity data corresponding to each dynamic window. Segmented processing can effectively reduce data redundancy and improve the real-time and accuracy of positioning. At the same time, the wireless signal receiver in the explosion-proof intelligent terminal collects signal strength data from surrounding WiFi access points and UWB base stations. The wireless signal receiver can detect the signal strength of various surrounding signal sources, and estimate the distance between the terminal and each access point or base station by analyzing the changes in these signals. Generally, the signal strength and distance show a certain attenuation relationship, which can be estimated by the following formula:

[0053] ;

[0054] in, Indicates the distance between the terminal and the signal source. is the transmission power of the signal source, is the signal strength received by the terminal, is the path loss factor. The relative distance between the terminal and multiple surrounding signal sources at each time point is calculated according to the above formula through the collected signal strength, and a wireless signal strength data set is constructed. The original acceleration data, original angular velocity data, segmented geomagnetic field strength data, and wireless signal strength data set are synchronized and aligned. Through interpolation or timestamp alignment algorithms, the data collected by different sensors in the same time period are associated to ensure that they reflect the same state of the terminal at the same time. Through time synchronization and data alignment, multi-source information from different sensors is integrated to generate a complete original sensor data set.

[0055] In a specific embodiment, the process of executing step S2 may specifically include the following steps:

[0056] Perform wavelet threshold denoising on the original acceleration data to obtain denoised acceleration data, and calculate the gravity component based on the denoised acceleration data to obtain linear acceleration data;

[0057] Filtering the original angular velocity data to obtain filtered angular velocity data, and performing attitude calculation based on the linear acceleration data and the filtered angular velocity data to obtain an attitude quaternion;

[0058] Based on the ellipsoid fitting algorithm, the segmented geomagnetic field intensity data is calibrated with soft iron and hard iron to obtain the calibrated geomagnetic field data;

[0059] Based on the calibrated geomagnetic field data and attitude quaternion, the projection of the geomagnetic field in the geographic coordinate system is calculated through the coordinate system conversion algorithm to obtain the geomagnetic north vector;

[0060] Remove outliers from the wireless signal strength data set, and convert signal strength to distance using a distance attenuation model to obtain wireless signal distance data;

[0061] The linear acceleration data, filtered angular velocity data, geomagnetic north vector and wireless signal distance data are divided and aligned to obtain inertial measurement unit data, geomagnetic field sensor data and wireless signal receiver data.

[0062] Specifically, the original acceleration data is subjected to wavelet transform, the signal is decomposed into different frequency components, and different thresholds are applied to each component to filter out noise. Through wavelet threshold denoising, useful information in the acceleration data is retained while high-frequency noise is eliminated. The process of wavelet threshold denoising can be performed by decomposing the original acceleration data and then applying appropriate thresholds in each frequency band to remove noise. For example, assuming the acceleration data is , decompose it into multiple components through wavelet transform, where the low-frequency part retains the main signal components, while the high-frequency part is mostly noise. Truncating or reducing the high-frequency components can effectively eliminate noise and obtain denoised acceleration data. Based on the denoised acceleration data, the gravity component acting on the terminal is calculated by analyzing the change trend of the acceleration. Usually, the gravity component can be extracted by low-pass filtering or directly from the static state of the acceleration data. Gravity component It represents the influence of the earth's gravity on the terminal in the z-axis direction. By removing the gravity component, linear acceleration data is obtained. These data only reflect the movement acceleration of the terminal, but do not include the influence of gravity. The original angular velocity data is filtered to eliminate the noise introduced by sensor errors or environmental interference. Common filtering techniques include low-pass filters and Kalman filters. Attitude solution is performed based on linear acceleration data and filtered angular velocity data. The attitude solution process usually uses the quaternion method, which can accurately describe the attitude changes of the terminal in three-dimensional space. Attitude quaternion The calculation needs to take into account the changes in linear acceleration and angular velocity at the same time to ensure the stability and accuracy of the attitude data. For the geomagnetic field data, soft iron and hard iron calibration are performed based on the ellipsoid fitting algorithm. During the measurement process, the geomagnetic field sensor is often interfered by soft iron and hard iron in the environment. These interferences will cause deviations in the magnetic field data. The soft iron effect will change the direction of the magnetic field, while the hard iron effect will introduce a constant offset. Through the ellipsoid fitting algorithm, the geomagnetic field data is fitted to an ideal ellipsoid surface to eliminate these interferences and obtain the calibrated geomagnetic field data. Based on the calibrated geomagnetic field data and attitude quaternion, the coordinate system conversion algorithm is used to calculate the projection of the geomagnetic field in the geographic coordinate system to obtain the geomagnetic north vector. Specifically, the attitude quaternion The geomagnetic field data is converted from the terminal coordinate system to the geographic coordinate system, expressed as . The geomagnetic north vector reflects the current direction of the terminal toward the geomagnetic north. For the wireless signal strength data set, outliers are removed. Due to obstacles or interference in the environment, the received signal strength may fluctuate abnormally, and these outliers will cause positioning errors. These outliers can be effectively removed through statistical methods such as median filtering or standard deviation analysis. The signal strength is converted into distance data based on the distance attenuation model. The relationship between signal strength and distance can usually be expressed by the following formula:

[0063] ;

[0064] in, Indicates the distance between the terminal and the signal source. is the transmission power of the signal source, is the signal strength received by the terminal, is the path loss factor. According to this formula, the received signal strength Estimate the distance between the terminal and each signal source to obtain the wireless signal distance data. Divide and align the linear acceleration data, filtered angular velocity data, geomagnetic north vector, and wireless signal distance data. Data division is to reasonably segment in the time dimension for subsequent analysis. Data alignment uses a time synchronization algorithm to ensure that data from different sensors can match at the same time point. Through interpolation or resampling methods, data of different frequencies are unified to the same time scale to generate complete inertial measurement unit data, geomagnetic field sensor data, and wireless signal receiver data.

[0065] In a specific embodiment, the process of executing step S3 may specifically include the following steps:

[0066] Perform coordinate system conversion on the linear acceleration data in the inertial measurement unit data to obtain acceleration data in the geographic coordinate system;

[0067] Based on the acceleration data in the geographic coordinate system, dynamic threshold analysis is performed to obtain the gait cycle detection results;

[0068] The acceleration data in the geographic coordinate system is segmented according to the gait cycle detection results to obtain single-step acceleration data;

[0069] Adaptively update the single-step acceleration data to zero speed to obtain the corrected acceleration data;

[0070] Performing a first integral operation based on the corrected acceleration data to obtain velocity data, and performing drift error correction on the velocity data to obtain corrected velocity data;

[0071] Perform a second integration operation based on the corrected velocity data to obtain displacement data;

[0072] The displacement data is accumulated and the coordinate transformation is performed in combination with the attitude quaternion in the inertial measurement unit data to obtain the initial relative displacement data of the explosion-proof intelligent terminal.

[0073] Specifically, the attitude is solved by using the gyroscope data and acceleration data in the inertial measurement unit to obtain the attitude information of the explosion-proof intelligent terminal. The acceleration data collected by the inertial measurement unit is usually expressed in the local coordinate system of the device itself, and in order to perform accurate displacement calculation, the acceleration data is converted to the geographic coordinate system. , the acceleration data in the local coordinate system Convert acceleration data into geographic coordinate system , the specific conversion formula is:

[0074] ;

[0075] in, is the attitude represented by the quaternion, is its inverse quaternion, is the acceleration vector in the local coordinate system, It is the acceleration vector in the geographic coordinate system. Through conversion, the motion data of the device is unified into the global coordinate system for processing. Dynamic threshold analysis is performed based on the acceleration data in the geographic coordinate system to detect the gait cycle and identify when each step of the user starts and ends. Usually, when people walk, the acceleration will show periodic fluctuations. Dynamic threshold analysis determines the cycle of the steps by tracking these fluctuations and detecting specific peaks and valleys. Assume that the acceleration component along the vertical direction is It has obvious periodic changes, and a dynamic threshold can be defined to detect the beginning and end of each gait. The acceleration data in the geographic coordinate system is segmented according to the gait cycle to form single-step acceleration data. Each segment of data corresponds to one step of the user. Each segment of single-step acceleration data is adaptively updated to zero speed, and the speed at that moment is forced to be updated to zero by identifying the user's stationary state at the end of each step. Because the foot of a person will contact the ground and stay for a short time at the end of each step, the speed should be zero. Through adaptive zero-speed update, the accumulated error of the acceleration sensor is effectively eliminated to obtain the corrected acceleration data. Perform the first integral operation on the corrected acceleration data to calculate the speed data. Acceleration is the rate of change of speed, so the speed The acceleration can be Integrate to get:

[0076] ;

[0077] Drift error correction is performed on the speed. Drift error can be corrected by analyzing the speed deviation during zero speed update. By forcibly correcting the speed to zero at each zero speed point in the gait cycle, the accumulated error in the speed is effectively reduced, and the corrected speed data is obtained. Based on the corrected velocity data, a second integral operation is performed to calculate the terminal displacement data. Since velocity is the rate of change of displacement, displacement It can be obtained by integrating the velocity:

[0078] ;

[0079] The displacement data represents the distance the user moves in each gait cycle. The displacement data is accumulated and combined with the attitude quaternion in the inertial measurement unit to perform coordinate transformation to improve the accuracy of displacement estimation. Attitude quaternion It is used to convert the displacement data from the local coordinate system to the global geographic coordinate system, so as to ensure that the calculated displacement direction is consistent with the actual movement direction of the user, and obtain the initial relative displacement data of the explosion-proof intelligent terminal.

[0080] In a specific embodiment, the process of executing step S4 may specifically include the following steps:

[0081] Initialize and calibrate the displacement sensor in the explosion-proof intelligent terminal to obtain the initial parameters of the sensor;

[0082] The original data collected by the displacement sensor is linearized based on the initial parameters of the sensor to obtain linearized displacement data;

[0083] Performing adaptive median filtering on the linearized displacement data to obtain filtered displacement data, and performing Fourier transform based on the filtered displacement data to obtain frequency domain feature data;

[0084] Perform peak detection and phase analysis on the frequency domain feature data to obtain the precise displacement, and perform coordinate transformation based on the precise displacement and a preset fixed reference point to obtain relative position data;

[0085] Performing noise suppression on the relative position data to obtain filtered relative position data;

[0086] The coordinate system is aligned based on the filtered relative position data and the posture information of the explosion-proof intelligent terminal to obtain the absolute displacement measurement data.

[0087] Specifically, the displacement sensor in the explosion-proof intelligent terminal is initialized and calibrated, including the calibration of the sensor's zero offset, sensitivity, and linear response. The initial parameters of the displacement sensor determine how the sensor converts physical displacement into electrical signals and how it responds to environmental changes. By calibrating the sensor in a known displacement environment, the relationship between the sensor's output and the actual displacement is measured to obtain the initial parameters of the sensor. The raw data collected by the sensor is linearized based on the initial parameters of the sensor, and the nonlinear response of the sensor is converted into linear data that is proportional to the actual displacement. In the initial data of the sensor, there may be nonlinear distortion introduced by hardware characteristics or environmental factors. Assume that the raw data output by the sensor is , which can be converted into linear displacement data after linearization , this conversion can be achieved through the linear relationship:

[0088] ;

[0089] in, is the proportionality constant of the sensor, is the zero offset measured during calibration. The function of this formula is to convert the original electrical signal Converted into actual displacement , eliminating fixed offset and proportional errors in the system. Adaptive median filtering is performed on the linearized displacement data. Adaptive median filtering is an effective denoising method suitable for processing signals containing impulse noise. By sorting the neighborhood data of each data point and taking the median, extreme value noise is effectively filtered out while retaining the true displacement change. The filtered displacement data is smoother, reducing the impact of environmental interference on the measurement results. Displacement data after filtering Reflects the terminal at time The actual displacement on the surface. Based on the filtered displacement data, Fourier transform is performed to convert the signal from the time domain to the frequency domain. The purpose of Fourier transform is to analyze the frequency components of the displacement signal and find out the hidden periodic characteristics and noise frequency distribution. The Fourier transform formula is:

[0090] ;

[0091] in, represents the frequency domain characteristics, is the displacement data after filtering, is the frequency. Through Fourier transform, the amplitude and phase of the signal at different frequencies are obtained to form frequency domain feature data. The results of frequency domain analysis help to understand the periodic changes in displacement data and identify possible noise frequencies or system errors in sensor signals. Peak detection and phase analysis are performed on the obtained frequency domain feature data to extract the precise displacement. Peak detection is used to identify significant peaks in frequency domain features, which correspond to the main components in the displacement signal. Phase analysis is used to determine the timing characteristics of the signal, especially in periodic motion, where phase information is particularly critical for determining the direction of motion and change trends. Based on the comprehensive analysis of peak and phase, the precise displacement is calculated. The coordinate transformation is performed based on the calculated precise displacement and the preset fixed reference point to obtain the relative position data. Assume that the coordinates of the fixed reference point are The system obtains the relative position of the terminal by superimposing the displacement onto the reference point coordinates.

[0092] ;

[0093] This formula represents the spatial position of the terminal relative to the initial fixed reference point, ensuring that the displacement data at each moment can match the actual physical space. In order to improve positioning accuracy, noise suppression is performed on the relative position data. Noise suppression can be achieved through filtering, smoothing and other methods to eliminate high-frequency interference, retain the main displacement change trends, and obtain smoother and more accurate relative position data. Through noise suppression, the subtle movement of the terminal in space is captured to reduce the accumulation of errors. Coordinate system alignment is performed based on the filtered relative position data and the attitude information of the explosion-proof intelligent terminal. Since the terminal may rotate or tilt in space, attitude information is the key to ensuring that the displacement data is consistent with the actual spatial position. Through attitude quaternion , convert the relative position data from the device coordinate system to the global coordinate system, ensuring that all displacement data are accurately calculated in the geographic coordinate system. Obtain the absolute displacement measurement data of the terminal .

[0094] In a specific embodiment, the process of executing step S5 may specifically include the following steps:

[0095] Performing coordinate system conversion and time synchronization on the initial relative displacement data and the absolute displacement measurement data to obtain preprocessed relative displacement data and preprocessed absolute displacement data;

[0096] Inputting the preprocessed relative displacement data and the preprocessed absolute displacement data into a multi-level Kalman filter model, wherein the multi-level Kalman filter model includes a plurality of levels, and the plurality of levels are a position estimation layer, a velocity estimation layer, and an acceleration estimation layer;

[0097] In the position estimation layer, the preprocessed relative displacement data is input as the state vector, and the preprocessed absolute displacement data is input as the observation vector, and the state prediction is performed through the first state prediction module of the position estimation layer to obtain the position prediction value; in the first measurement update module of the position estimation layer, the first Kalman gain is calculated based on the position prediction value and the position observation value; in the first state fusion module of the position estimation layer, the position prediction value is corrected based on the first Kalman gain to obtain the position state estimation value;

[0098] In the velocity estimation layer, the preprocessed relative displacement data is subjected to first-order difference to obtain velocity data, and the velocity data is used as a state vector input. The state prediction module of the velocity estimation layer performs state prediction to obtain a velocity prediction value; in the second measurement update module of the velocity estimation layer, the second Kalman gain is calculated based on the velocity prediction value and the velocity observation value; in the second state fusion module of the velocity estimation layer, the velocity prediction value is corrected based on the second Kalman gain to obtain a velocity state estimation value;

[0099] In the acceleration estimation layer, the velocity data is first-order differentiated to obtain acceleration data, the acceleration data is used as the state vector input, and the state prediction is performed through the state prediction module of the acceleration estimation layer to obtain the acceleration prediction value; in the third measurement update module of the acceleration estimation layer, the third Kalman gain is calculated based on the acceleration prediction value and the acceleration observation value; in the third state fusion module of the acceleration estimation layer, the acceleration prediction value is corrected based on the third Kalman gain to obtain the acceleration state estimation value;

[0100] The position state estimation value, velocity state estimation value and acceleration state estimation value are weightedly fused to obtain the terminal position data.

[0101] Specifically, the initial relative displacement data and the absolute displacement measurement data are converted to coordinate systems and synchronized in time. The relative displacement data of the inertial measurement unit is usually obtained in the device coordinate system, while the absolute displacement measurement data may come from external sensors and is usually expressed in the geographic coordinate system. In order to ensure data consistency, the relative displacement data is converted from the device coordinate system to the geographic coordinate system through the attitude quaternion, which can be achieved by the following formula:

[0102] ;

[0103] in, Represents the relative displacement data converted to the geographic coordinate system. is the attitude quaternion, is the original relative displacement data, is the inverse of the quaternion. Through the conversion steps, all displacement data are unified into the geographic coordinate system to ensure that they can be effectively compared. In terms of time synchronization, the data of all sensors may not be at the same timestamp due to different sampling frequencies. Use the interpolation algorithm to synchronize all data to the same time base. Through interpolation or resampling technology, ensure that the initial relative displacement data and absolute displacement measurement data can be compared at the same time point to obtain the preprocessed relative displacement data and absolute displacement data. Input the preprocessed relative displacement data and absolute displacement data into the multi-level Kalman filter model. The multi-level Kalman filter model consists of three levels: position estimation layer, velocity estimation layer, and acceleration estimation layer. Each level processes the corresponding data separately to help gradually reduce noise and accumulated errors and improve the accuracy of displacement estimation. In the position estimation layer, the preprocessed relative displacement data is input as the state vector, and the preprocessed absolute displacement data is input as the observation vector into the Kalman filter model. Through the first state prediction module of the position estimation layer, the position prediction value is obtained based on the current state prediction. The state prediction formula of the Kalman filter is as follows:

[0104] ;

[0105] in, For the current moment The predicted value of the position, is the state transfer matrix, which represents the process of the system transferring from the state of the previous time step to the current time step. is the estimated position at the previous moment, is the control input matrix, is the control variable. After the position is predicted, in the first measurement update module, the first Kalman gain is calculated based on the position prediction value and the observed position value. Kalman gain It is used to make a weighted balance between state prediction and observation. The formula is:

[0106] ;

[0107] in, is the prior error covariance matrix, is the observation matrix, is the observation noise covariance matrix. The predicted position is corrected by the Kalman gain. In the first state fusion module, the first Kalman gain is used to correct the position prediction value to obtain the final position state estimate. In the velocity estimation layer, the velocity data is obtained by performing a first-order difference calculation on the preprocessed relative displacement data. The velocity is the time rate of change of the displacement, and the formula is:

[0108] ;

[0109] in, is the speed at the current moment, is the displacement at the current moment, is the time interval between two time points. After obtaining the velocity data, the velocity is input as the state vector into the state prediction module of the velocity estimation layer, and the velocity prediction is performed in a similar manner to the position estimation layer to obtain the velocity prediction value. Similarly, the second measurement update module of the velocity estimation layer calculates the second Kalman gain based on the velocity prediction value and the actual observed velocity, and in the second state fusion module, the velocity prediction value is corrected based on the gain to obtain the velocity state estimation value. In the acceleration estimation layer, the velocity data is first-order differentiated to calculate the acceleration data. Acceleration is the time rate of change of velocity, and the formula is:

[0110] ;

[0111] in, is the acceleration at the current moment, is the velocity value. The acceleration data is input as the state vector into the acceleration estimation layer to predict the acceleration through the state prediction module. Similarly, in the third measurement update module, the third Kalman gain is calculated based on the acceleration prediction value and the observation value, and in the third state fusion module, the acceleration prediction value is corrected using the gain to finally obtain the acceleration state estimation value. The position state estimation value, velocity state estimation value and acceleration state estimation value are weightedly fused to improve the overall displacement estimation accuracy. Through comprehensive processing of data at different levels, the consistency between position, velocity and acceleration is ensured to obtain the final terminal position data.

[0112] In a specific embodiment, the execution steps include: in a position estimation layer, the preprocessed relative displacement data is input as a state vector, the preprocessed absolute displacement data is input as an observation vector, and a state prediction is performed through a first state prediction module of the position estimation layer to obtain a position prediction value; in a first measurement update module of the position estimation layer, a first Kalman gain is calculated based on the position prediction value and the position observation value; in a first state fusion module of the position estimation layer, the position prediction value is corrected based on the first Kalman gain, and the process of obtaining a position state estimation value can specifically include the following steps:

[0113] In the first state prediction module of the position estimation layer, the state vector is initialized for the preprocessed relative displacement data to obtain an initial state vector and an initial covariance matrix;

[0114] Based on the initial state vector and the initial covariance matrix, the state is predicted through the state transfer matrix to obtain the prior state vector, and the covariance of the prior state vector is predicted based on the process noise covariance matrix to obtain the prior covariance matrix;

[0115] An observation vector is constructed based on the preprocessed absolute displacement data, and the prior state vector is mapped to the observation space through the observation matrix to obtain a predicted observation vector;

[0116] Calculate the residual of the predicted observation vector and the observation vector to obtain the observation residual;

[0117] In a first measurement update module of the position estimation layer, a first Kalman gain is calculated based on a priori covariance matrix, an observation matrix and an observation noise covariance matrix;

[0118] In the first state fusion module of the position estimation layer, the first Kalman gain and the observation residual are used to correct the prior state vector to obtain the posterior state vector. Based on the first Kalman gain, the observation matrix and the prior covariance matrix, the covariance matrix is ​​updated to obtain the posterior covariance matrix, where the posterior state vector is the position state estimate.

[0119] Specifically, the relative displacement data is represented as a state vector. Usually contains variables such as position and velocity, which together describe the state of the system at a certain point in time. The initialization process needs to generate the initial state vector based on the preprocessed relative displacement data , which can include the current position information and initial velocity, and may be in the form of:

[0120] ;

[0121] in, is the initial displacement, is the initial velocity. Initialize the covariance matrix , which represents the uncertainty of each variable in the state vector. The covariance matrix is ​​usually a diagonal matrix, and the initial diagonal elements represent the initial estimate of the uncertainty of position and velocity. After the state vector is initialized, based on the initial state vector and the initial covariance matrix, the state transfer matrix Perform state prediction. State transfer matrix Used to transfer the system from the current state to the next state, its form can depend on the physical model of the system. For example, assuming that the state vector includes position and velocity, the state transfer matrix can be expressed as:

[0122] ;

[0123] in, is the time step, which represents the time interval between two adjacent time points. The state prediction formula is:

[0124] ;

[0125] in, is the prior state vector, representing the current moment Based on the last moment This formula shows that the update of position depends on the previous speed, while the speed remains unchanged. At the same time, based on the process noise covariance matrix , predict the covariance of the prior state vector, the formula is:

[0126] ;

[0127] in, is the prior covariance matrix, indicating the The uncertainty of the state. The process noise covariance matrix Represents the influence of random noise in the system on state prediction. Construct observation vector based on preprocessed absolute displacement data , the observation vector represents the actual measurement value of the system at the current moment. Then through the observation matrix , maps the prior state vector to the observation space. The observation matrix The role of is to convert the variables in the state vector (such as position and velocity) into a form that can be directly compared with the observed values. Assuming that only the position is observed, the observation matrix It can be expressed as:

[0128] ;

[0129] The observation equation is:

[0130] ;

[0131] in, is the predicted observation vector, which represents the observation value obtained through state prediction. The observation residual is calculated, that is, the difference between the predicted observation vector and the actual observation vector, and the formula is:

[0132] ;

[0133] Observation residuals It reflects the deviation between the displacement predicted by the system and the actual measured absolute displacement. In the first measurement update module of the position estimation layer, based on the prior covariance matrix , observation matrix and the observation noise covariance matrix , calculate the Kalman gain The Kalman gain represents the coefficient of the system's weighted balance between state prediction and observation, and the calculation formula is:

[0134] ;

[0135] in, is the observation noise covariance matrix, which represents the random error introduced in the observation process. The purpose of calculating the Kalman gain is to properly balance the influence of the predicted value and the observed value in the state correction process. In the first state fusion module of the position estimation layer, the Kalman gain is used and the observed residuals Correct the prior state vector, the formula is:

[0136] ;

[0137] in, is the posterior state vector, which is the estimated result after correcting the prior state vector. Through this correction step, the predicted value is combined with the actual observed value to obtain a more accurate displacement estimate. At the same time, the covariance matrix also needs to be updated, the formula is:

[0138] ;

[0139] in, is the identity matrix, and the updated represents the posterior covariance matrix, which reflects the uncertainty of the corrected state. The updating process ensures that the covariance matrix can reflect the uncertainty level at the current moment when the next state is predicted. The posterior state vector It is the position state estimate, which combines the system's prediction and observation information to ultimately give the optimal displacement estimate.

[0140] In a specific embodiment, the process of executing step S6 may specifically include the following steps:

[0141] Perform time synchronization and data alignment on the terminal location data, geomagnetic field sensor data, and wireless signal receiver data to obtain aligned multi-source data;

[0142] Input the aligned multi-source data into the feature extraction layer of the multi-source data fusion model, extract features through the convolutional neural network, obtain the multi-source feature map, perform self-attention mechanism analysis on the multi-source feature map, calculate the correlation weights between features, and obtain the weighted feature map;

[0143] The weighted feature map is input into the fusion layer of the multi-source data fusion model, and the features from different sources are interacted and fused through the cross attention mechanism to obtain the fused features. The fused features are then subjected to residual connection and layer normalization to obtain the optimized fused features.

[0144] The optimized fusion features are input into the three-dimensional space mapping layer of the multi-source data fusion model, and the optimized fusion features are mapped to three-dimensional space coordinates through a fully connected neural network to obtain the initial three-dimensional position coordinates;

[0145] The initial three-dimensional position coordinates are spatially constrained and optimized through conditional random fields to obtain the optimized three-dimensional position coordinates. The optimized three-dimensional position coordinates are input into the Kalman smoother, and the positions of multiple time steps are smoothed to obtain the three-dimensional spatial positioning results of the explosion-proof intelligent terminal.

[0146] Specifically, the terminal location data, geomagnetic field sensor data, and wireless signal receiver data are synchronized and aligned. The time step of each sensor to collect data may be different. For example, the geomagnetic field sensor may sample at a lower frequency, while the inertial measurement unit (IMU) and wireless signal receiver data have a higher frequency. In order to ensure that these data are processed under the same time reference, they are synchronized by interpolation, resampling, or timestamp alignment. Assume there are three types of data: terminal location data , geomagnetic field sensor data and wireless signal receiver data , through interpolation , and Synchronize to the same timestamp set , and get the aligned multi-source data. The aligned multi-source data is input into the feature extraction layer of the multi-source data fusion model. The feature extraction layer processes the multi-source data through a convolutional neural network (CNN). The convolutional neural network can extract the spatial features of the data. Multi-source data such as position, geomagnetic field, and signal strength have different physical meanings, but through the convolution operation, the system can automatically identify the hidden feature patterns in each data and merge these features into a multi-source feature map. The convolution operation scans the data through a sliding window to capture the local spatial correlation in different data sources. The generated multi-source feature map can simultaneously reflect the correlation between position, magnetic field strength, and signal strength. Assume that the convolution kernel is , the input multi-source data is , the convolution operation can be expressed as:

[0147] ;

[0148] Among them, * represents the convolution operation, and the generated feature map contains spatial features at different levels. After extracting the multi-source feature map, the self-attention mechanism is used for analysis. The self-attention mechanism can automatically calculate the correlation weights between different features. The main function of the self-attention mechanism is to establish connections between different parts of the data and identify which features are more important for the positioning task. By calculating the similarity between each feature and other features, a weight is assigned to each feature, and these weights determine the importance of the feature. The core formula of the self-attention mechanism is:

[0149] ;

[0150] in, , and are query, key, and value matrices respectively, is the dimension of the key, and the softmax function is used to normalize the weights. Through this process, a weighted feature map is generated, in which the contribution of each feature is re-evaluated by the self-attention mechanism. The weighted feature map is input to the fusion layer of the multi-source data fusion model. The fusion layer uses the cross-attention mechanism to process features from different sources to achieve deep interaction of data. The cross-attention mechanism allows data from different sensors to influence each other, thereby combining the advantages of each data source. The role of this mechanism is to ensure that the fusion process of multi-source data not only takes into account their respective features, but also improves the final positioning accuracy through interaction. After the cross-attention mechanism, the fused features are optimized through residual connections and layer normalization. The purpose of residual connections is to prevent the loss of information in deep networks. Even after multiple layers of processing, the original input features can still be retained through residual connections. Layer normalization is to ensure that the data distribution at different levels of the network is stable to avoid the problem of gradient disappearance or explosion. The optimized fused features are used for subsequent positioning calculations. The optimized fused features are input to the 3D spatial mapping layer of the multi-source data fusion model. In this layer, the optimized fusion features are mapped to the three-dimensional space coordinate system through the fully connected neural network to generate the initial three-dimensional position coordinates. The function of the fully connected layer is to convert the high-dimensional fusion features into specific three-dimensional coordinate values, that is, the position of the terminal in space. This process can be represented by the linear transformation of the fully connected neural network:

[0151] ;

[0152] in, is the weight matrix, is the bias term, is the input fusion feature, is the output three-dimensional coordinate. The initial three-dimensional position coordinates are optimized by spatial constraints through conditional random fields. Conditional random fields are a probabilistic model that can combine the position relationship between multiple time steps and add constraints to the position information of each time step to improve the continuity and accuracy of positioning. Conditional random fields optimize position estimation by maximizing the joint probability distribution of position sequences. The formula is:

[0153] ;

[0154] in, represents the relationship between the position and the input features, represents the interaction between different positions, is a normalization constant. By optimizing this joint probability, more accurate three-dimensional position coordinates are obtained. The optimized three-dimensional position coordinates are input into the Kalman smoother to smooth the position trajectory between multiple time steps to eliminate short-term noise or error fluctuations. The Kalman smoother combines forward prediction and smoothing steps to correct the estimate of each time point through the global information of the time series. After smoothing, the three-dimensional spatial positioning result of the explosion-proof intelligent terminal is obtained.

[0155] In a specific embodiment, the step of inputting the optimized fusion features into the three-dimensional space mapping layer of the multi-source data fusion model, mapping the optimized fusion features to the three-dimensional space coordinates through the fully connected neural network, and obtaining the initial three-dimensional position coordinates may specifically include the following steps:

[0156] The optimized fusion features are input into the first fully connected sublayer of the three-dimensional space mapping layer, and nonlinear transformation is performed through the ReLU activation function to obtain the first hidden layer output, and the first hidden layer output is batch normalized to obtain the standardized first hidden layer output;

[0157] The standardized first hidden layer output is input into the second fully connected sublayer of the three-dimensional space mapping layer, and features are extracted through the LeakyReLU activation function to obtain the second hidden layer output, and the second hidden layer output is dropout regularized to obtain the regularized second hidden layer output;

[0158] The regularized output of the second hidden layer is input into the third fully connected sublayer of the three-dimensional space mapping layer, and feature mapping is performed through the Tanh activation function to obtain the output of the third hidden layer;

[0159] Perform residual connection on the output of the third hidden layer, add the output of the third hidden layer to the optimized fusion feature, and obtain the residual enhanced feature representation;

[0160] The feature representation of the residual enhancement is input into the output layer of the three-dimensional space mapping layer, and the feature is mapped to the three-dimensional space coordinate system through linear transformation to obtain the initial three-dimensional position coordinates.

[0161] Specifically, the optimized fusion feature is input into the first fully connected sublayer of the three-dimensional space mapping layer to map the input high-dimensional features to the output space. Assume that the optimized fusion feature vector is , after the linear transformation of the first fully connected sublayer, the first hidden layer output is generated . The process can be expressed as:

[0162] ;

[0163] in, is the weight matrix of the first fully connected sublayer, is the bias vector, is the linear output of the first hidden layer. The ReLU (Rectified Linear Unit) activation function is applied to this output. ReLU is a commonly used nonlinear transformation function defined as:

[0164] ;

[0165] The main function of ReLU is to introduce nonlinearity, so that the model can learn complex mapping relationships, and at the same time has simple gradient calculation characteristics. The ReLU activation function will reset all negative values ​​to zero, retain positive values, and obtain the nonlinear output of the first hidden layer. In order to stabilize the training process of the network, the output of the first hidden layer is batch normalized. Batch normalization standardizes the output of each layer to have zero mean and unit variance to prevent problems such as gradient vanishing or gradient exploding. The formula for batch normalization is:

[0166] ;

[0167] in, is the sample value in the batch, and are the mean and variance of the batch samples, respectively. is to avoid the minimum constant of division by zero. Through batch normalization, the standardized output of the first hidden layer is obtained to enhance the robustness of the network. The standardized output of the first hidden layer is input to the second fully connected sublayer of the three-dimensional spatial mapping layer. Similar to the first layer, the second fully connected sublayer performs a linear transformation on the input data:

[0168] ;

[0169] in, is the weight matrix of the second layer, is the bias term, is the output of the first hidden layer after batch normalization. After completing the linear transformation, the Leaky ReLU activation function is used for feature extraction. Leaky ReLU is a variant of ReLU and is defined as:

[0170] ;

[0171] in, is a very small positive number, usually 0.01. Leaky ReLU allows a small part of negative values ​​to pass through, so that the network can also learn in the negative area. Through the activation of Leaky ReLU, the nonlinear output of the second hidden layer is obtained. In order to prevent overfitting problems, dropout regularization is introduced after the output of the second hidden layer. Dropout reduces the model's dependence on specific features by randomly discarding a part of neurons, thereby enhancing the model's generalization ability. Assume that the retention probability of dropout is , then the formula for dropout regularization is:

[0172] ;

[0173] in, is a A mask vector of the same dimension, whose elements are expressed in terms of probabilities is 1, The probability of is 0. By randomly discarding some neurons, the regularized output of the second hidden layer is obtained. The regularized output of the second hidden layer is input to the third fully connected sublayer of the three-dimensional space mapping layer to perform a linear transformation:

[0174] ;

[0175] in, is the weight matrix of the third layer, is the bias term. The third hidden layer output is feature mapped by the Tanh activation function. The Tanh activation function is defined as:

[0176] ;

[0177] Tanh compresses the input data to the interval [-1, 1], which can handle symmetry problems, so that the output features contain both positive and negative values, and the output of the third hidden layer is obtained. In order to ensure that the model can effectively retain the feature information of the previous layers, residual connections are performed to directly add the input of the previous layer to the output of the current layer to ensure that the original information is not lost in the deep network. Output of the third hidden layer With the optimized fusion features Add together to get the feature representation of residual enhancement:

[0178] ;

[0179] Ensure that even after multiple layers of nonlinear transformations, the model can still retain the original fusion feature information to avoid information attenuation in the deep network. The feature representation of the residual enhancement is input to the output layer of the three-dimensional space mapping layer. The output layer is a fully connected layer that maps the residual enhanced features to three-dimensional space coordinates. Through linear transformation, the initial three-dimensional position coordinates are generated:

[0180] ;

[0181] in, is the weight matrix of the output layer, is the bias term, is the initial three-dimensional position coordinate. This coordinate represents the position of the explosion-proof intelligent terminal in three-dimensional space.

[0182] The above describes the explosion-proof intelligent terminal positioning method based on the high-end displacement sensor in the embodiment of the present invention. The following describes the explosion-proof intelligent terminal positioning device based on the high-end displacement sensor in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an explosion-proof intelligent terminal positioning device based on a high-end displacement sensor includes:

[0183] The acquisition module is used to collect data from the inertial measurement unit, geomagnetic field sensor and wireless signal receiver in the explosion-proof intelligent terminal to obtain the original sensor data set;

[0184] A calibration module is used to filter and calibrate the original sensor data set to obtain inertial measurement unit data, geomagnetic field sensor data and wireless signal receiver data;

[0185] A calculation module, used for performing double integral calculation based on the inertial measurement unit data to obtain the initial relative displacement data of the explosion-proof intelligent terminal;

[0186] A measurement module is used to measure the position of the explosion-proof intelligent terminal relative to a fixed reference point through a displacement sensor to obtain absolute displacement measurement data;

[0187] An analysis module is used to input the initial relative displacement data and the absolute displacement measurement data into a multi-level Kalman filter model for position analysis to obtain terminal position data;

[0188] The positioning module is used to input the terminal location data, geomagnetic field sensor data and wireless signal receiver data into the multi-source data fusion model for three-dimensional spatial positioning to obtain the three-dimensional spatial positioning result of the explosion-proof intelligent terminal.

[0189] Through the synergy of the above components, the data of the inertial measurement unit, geomagnetic field sensor, wireless signal receiver and high-end displacement sensor are combined, and the multi-source data fusion model is used for comprehensive analysis, which effectively improves the accuracy and robustness of positioning. The introduction of high-end displacement sensors for absolute displacement measurement provides an accurate reference point for the inertial navigation system, effectively suppresses the cumulative error, and significantly improves the accuracy of long-term positioning. The use of dynamic window division, adaptive zero-speed update and other technologies can automatically adjust the data processing strategy according to the motion state of the terminal, improving the adaptability and accuracy of the algorithm. A multi-level Kalman filter model including position, velocity and acceleration estimation layers is designed, and more stable and accurate state estimation is achieved through inter-layer information transmission and correction. The introduction of convolutional neural network and attention mechanism in the multi-source data fusion model can automatically learn and extract the key features of multi-source data, improving the fusion effect. The spatial constraint optimization of the initial three-dimensional position coordinates is carried out through conditional random fields, and the rationality of the positioning results is further improved by combining environmental prior knowledge. The neural network structure is optimized by batch normalization, dropout and other technologies, which reduces the computational complexity while ensuring the positioning accuracy, and is suitable for running on resource-constrained explosion-proof intelligent terminals. The present invention takes into account the special requirements of explosion-proof environments, avoids the problems of high power consumption and heating, and ensures safe application in hazardous environments.

[0190] The present invention also provides a computer device, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the explosion-proof intelligent terminal positioning method based on a high-end displacement sensor in the above-mentioned embodiments.

[0191] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the explosion-proof intelligent terminal positioning method based on a high-end displacement sensor.

[0192] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0193] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0194] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for positioning an explosion-proof intelligent terminal based on a high-end displacement sensor, characterized in that: The method comprises: Collect data from the inertial measurement unit, geomagnetic field sensor and wireless signal receiver in the explosion-proof intelligent terminal to obtain the original sensor data set; Filtering and calibrating the raw sensor data set to obtain inertial measurement unit data, geomagnetic field sensor data, and wireless signal receiver data; Performing a double integral operation based on the inertial measurement unit data to obtain initial relative displacement data of the explosion-proof intelligent terminal; The position of the explosion-proof intelligent terminal relative to a fixed reference point is measured by a displacement sensor to obtain absolute displacement measurement data; The initial relative displacement data and the absolute displacement measurement data are input into a multi-level Kalman filter model for position analysis to obtain terminal position data; specifically comprising: performing coordinate system conversion and time synchronization on the initial relative displacement data and the absolute displacement measurement data to obtain pre-processed relative displacement data and pre-processed absolute displacement data; inputting the pre-processed relative displacement data and the pre-processed absolute displacement data into a multi-level Kalman filter model, the multi-level Kalman filter model comprising a plurality of levels, the plurality of levels being a position estimation layer, a velocity estimation layer and an acceleration estimation layer; in the position estimation layer, the pre-processed relative displacement data is input as a state vector, the pre-processed absolute displacement data is input as an observation vector, and a state prediction is performed through a first state prediction module of the position estimation layer to obtain a position prediction value; in a first measurement update module of the position estimation layer, a first Kalman gain is calculated based on the position prediction value and the position observation value; in a first state fusion module of the position estimation layer, the position prediction value is corrected based on the first Kalman gain to obtain a position state estimation value; in a velocity estimation layer, The pre-processed relative displacement data is subjected to first-order difference to obtain velocity data, and the velocity data is input as a state vector, and a state prediction is performed through a state prediction module of a velocity estimation layer to obtain a velocity prediction value; in a second measurement update module of the velocity estimation layer, a second Kalman gain is calculated based on the velocity prediction value and the velocity observation value; in a second state fusion module of the velocity estimation layer, the velocity prediction value is corrected based on the second Kalman gain to obtain a velocity state estimation value; in an acceleration estimation layer, a first-order difference is performed on the velocity data to obtain acceleration data, and the acceleration data is input as a state vector, and a state prediction is performed through a state prediction module of the acceleration estimation layer to obtain an acceleration prediction value; in a third measurement update module of the acceleration estimation layer, a third Kalman gain is calculated based on the acceleration prediction value and the acceleration observation value; in a third state fusion module of the acceleration estimation layer, the acceleration prediction value is corrected based on the third Kalman gain to obtain an acceleration state estimation value; weighted fusion is performed on the position state estimation value, the velocity state estimation value and the acceleration state estimation value to obtain terminal position data; Input the terminal location data, the geomagnetic field sensor data and the wireless signal receiver data into a multi-source data fusion model for three-dimensional spatial positioning, and obtain the three-dimensional spatial positioning result of the explosion-proof intelligent terminal; specifically including: performing time synchronization and data alignment on the terminal location data, the geomagnetic field sensor data and the wireless signal receiver data to obtain aligned multi-source data; input the aligned multi-source data into the feature extraction layer of the multi-source data fusion model, perform feature extraction through a convolutional neural network to obtain a multi-source feature map, and perform self-attention mechanism analysis on the multi-source feature map, calculate the correlation weights between features, and obtain a weighted feature map; The weighted feature map is input into the fusion layer of the multi-source data fusion model, and the features from different sources are interacted and fused through the cross attention mechanism to obtain fusion features, and the fusion features are subjected to residual connection and layer normalization processing to obtain optimized fusion features; the optimized fusion features are input into the first fully connected sublayer of the three-dimensional space mapping layer, and nonlinear transformation is performed through the ReLU activation function to obtain the first hidden layer output, and the first hidden layer output is subjected to batch normalization processing to obtain a standardized first hidden layer output; the standardized first hidden layer output is input into the second fully connected sublayer of the three-dimensional space mapping layer, and the first hidden layer output is subjected to the Leaky The ReLU activation function is used to extract features to obtain the output of the second hidden layer, and the second hidden layer output is subjected to dropout regularization to obtain the regularized second hidden layer output; the regularized second hidden layer output is input into the third fully connected sublayer of the three-dimensional space mapping layer, and feature mapping is performed through the Tanh activation function to obtain the output of the third hidden layer; the third hidden layer output is subjected to residual connection, and the third hidden layer output is added to the optimized fusion feature to obtain the residual enhanced feature representation; the residual enhanced feature representation is input into the output layer of the three-dimensional space mapping layer, and the feature is mapped to the three-dimensional space coordinate system through linear transformation to obtain the initial three-dimensional position coordinates; the initial three-dimensional position coordinates are subjected to spatial constraint optimization through conditional random fields to obtain the optimized three-dimensional position coordinates, and the optimized three-dimensional position coordinates are input into the Kalman smoother to smooth the positions of multiple time steps to obtain the three-dimensional space positioning result of the explosion-proof intelligent terminal.

2. The explosion-proof intelligent terminal positioning method based on high-end displacement sensor according to claim 1 is characterized in that: The inertial measurement unit, the geomagnetic field sensor and the wireless signal receiver in the explosion-proof intelligent terminal are collected to obtain the original sensor data set, including: Perform high-frequency sampling on the accelerometer and gyroscope in the inertial measurement unit in the explosion-proof intelligent terminal to obtain raw acceleration data and raw angular velocity data; Performing zero velocity detection based on the original acceleration data and the original angular velocity data to obtain a stationary state mark sequence; Segmenting the original acceleration data and the original angular velocity data according to the stationary state mark sequence to obtain a dynamic window division result; Sampling the geomagnetic field sensor in the explosion-proof intelligent terminal and combining the dynamic window division result to obtain segmented geomagnetic field intensity data; Based on the wireless signal receiver in the explosion-proof intelligent terminal, the signal strength of the surrounding WiFi access points and UWB base stations is collected to obtain a wireless signal strength data set; The original acceleration data, the original angular velocity data, the segmented geomagnetic field strength data and the wireless signal strength data set are time synchronized and data aligned to obtain an original sensor data set.

3. The explosion-proof intelligent terminal positioning method based on high-end displacement sensor according to claim 2 is characterized in that: The filtering and calibration processing of the raw sensor data set to obtain inertial measurement unit data, geomagnetic field sensor data and wireless signal receiver data includes: Performing wavelet threshold denoising on the original acceleration data to obtain denoised acceleration data, and calculating the gravity component based on the denoised acceleration data to obtain linear acceleration data; Filtering the original angular velocity data to obtain filtered angular velocity data, and performing attitude calculation based on the linear acceleration data and the filtered angular velocity data to obtain an attitude quaternion; Performing soft iron and hard iron calibration on the segmented geomagnetic field intensity data based on an ellipsoid fitting algorithm to obtain calibrated geomagnetic field data; Based on the calibrated geomagnetic field data and the attitude quaternion, a projection of the geomagnetic field in the geographic coordinate system is calculated by a coordinate system conversion algorithm to obtain a geomagnetic north vector; Remove outliers from the wireless signal strength data set, and convert signal strength to distance using a distance attenuation model to obtain wireless signal distance data; The linear acceleration data, the filtered angular velocity data, the geomagnetic north vector and the wireless signal distance data are divided and aligned to obtain inertial measurement unit data, geomagnetic field sensor data and wireless signal receiver data.

4. The explosion-proof intelligent terminal positioning method based on high-end displacement sensor according to claim 3 is characterized in that: The double integration operation is performed based on the inertial measurement unit data to obtain the initial relative displacement data of the explosion-proof intelligent terminal, including: Performing coordinate system conversion on the linear acceleration data in the inertial measurement unit data to obtain acceleration data in a geographic coordinate system; Performing dynamic threshold analysis based on the acceleration data in the geographic coordinate system to obtain a gait cycle detection result; Segmenting the acceleration data in the geographic coordinate system according to the gait cycle detection result to obtain single-step acceleration data; Adaptively updating the single-step acceleration data at zero speed to obtain corrected acceleration data; Performing a first integral operation based on the corrected acceleration data to obtain velocity data, and performing drift error correction on the velocity data to obtain corrected velocity data; Perform a second integration operation based on the corrected velocity data to obtain displacement data; The displacement data is accumulated and coordinate transformed in combination with the attitude quaternion in the inertial measurement unit data to obtain the initial relative displacement data of the explosion-proof intelligent terminal.

5. The explosion-proof intelligent terminal positioning method based on high-end displacement sensor according to claim 1 is characterized in that: The method of measuring the position of the explosion-proof intelligent terminal relative to a fixed reference point by a displacement sensor to obtain absolute displacement measurement data includes: Initializing and calibrating the displacement sensor in the explosion-proof intelligent terminal to obtain initial parameters of the sensor; Performing linearization processing on the raw data collected by the displacement sensor based on the initial parameters of the sensor to obtain linearized displacement data; Performing adaptive median filtering on the linearized displacement data to obtain filtered displacement data, and performing Fourier transform based on the filtered displacement data to obtain frequency domain feature data; Performing peak detection and phase analysis on the frequency domain characteristic data to obtain precise displacement, and performing coordinate transformation based on the precise displacement and a preset fixed reference point to obtain relative position data; Performing noise suppression on the relative position data to obtain filtered relative position data; Coordinate system alignment is performed based on the filtered relative position data and the posture information of the explosion-proof intelligent terminal to obtain absolute displacement measurement data.

6. The explosion-proof intelligent terminal positioning method based on high-end displacement sensor according to claim 1 is characterized in that: In the position estimation layer, the preprocessed relative displacement data is input as a state vector, and the preprocessed absolute displacement data is input as an observation vector, and a state prediction is performed through a first state prediction module of the position estimation layer to obtain a position prediction value; in a first measurement update module of the position estimation layer, a first Kalman gain is calculated based on the position prediction value and the position observation value; In a first state fusion module of the position estimation layer, the position prediction value is corrected based on the first Kalman gain to obtain a position state estimation value, including: In the first state prediction module of the position estimation layer, initializing the state vector of the preprocessed relative displacement data to obtain an initial state vector and an initial covariance matrix; Based on the initial state vector and the initial covariance matrix, a state prediction is performed through a state transfer matrix to obtain a priori state vector, and covariance prediction is performed on the priori state vector based on a process noise covariance matrix to obtain a priori covariance matrix; Constructing an observation vector based on the preprocessed absolute displacement data, and mapping the prior state vector to the observation space through an observation matrix to obtain a predicted observation vector; Performing residual calculation on the predicted observation vector and the observation vector to obtain an observation residual; In a first measurement update module of the position estimation layer, a first Kalman gain is calculated based on the prior covariance matrix, the observation matrix and the observation noise covariance matrix; In the first state fusion module of the position estimation layer, the prior state vector is corrected using the first Kalman gain and the observation residual to obtain a posterior state vector, and the covariance matrix is ​​updated based on the first Kalman gain, the observation matrix and the prior covariance matrix to obtain a posterior covariance matrix, wherein the posterior state vector is the position state estimate.

7. An explosion-proof intelligent terminal positioning device based on a high-end displacement sensor, characterized in that: Used to execute the explosion-proof intelligent terminal positioning method based on a high-end displacement sensor according to any one of claims 1 to 6, the device comprises: The acquisition module is used to collect data from the inertial measurement unit, geomagnetic field sensor and wireless signal receiver in the explosion-proof intelligent terminal to obtain the original sensor data set; A calibration module, used for filtering and calibrating the raw sensor data set to obtain inertial measurement unit data, geomagnetic field sensor data and wireless signal receiver data; A calculation module, used for performing a double integral operation based on the inertial measurement unit data to obtain the initial relative displacement data of the explosion-proof intelligent terminal; A measuring module, used to measure the position of the explosion-proof intelligent terminal relative to a fixed reference point through a displacement sensor to obtain absolute displacement measurement data; The analysis module is used to input the initial relative displacement data and the absolute displacement measurement data into a multi-level Kalman filter model for position analysis to obtain terminal position data; specifically comprising: performing coordinate system conversion and time synchronization on the initial relative displacement data and the absolute displacement measurement data to obtain pre-processed relative displacement data and pre-processed absolute displacement data; inputting the pre-processed relative displacement data and the pre-processed absolute displacement data into a multi-level Kalman filter model, wherein the multi-level Kalman filter model includes multiple levels, and the multiple levels are a position estimation layer, a velocity estimation layer, and an acceleration estimation layer; in the position estimation layer, the pre-processed relative displacement data is input as a state vector, and the pre-processed absolute displacement data is input as an observation vector, and a state prediction is performed through a first state prediction module of the position estimation layer to obtain a position prediction value; in a first measurement update module of the position estimation layer, a first Kalman gain is calculated based on the position prediction value and the position observation value; in a first state fusion module of the position estimation layer, the position prediction value is corrected based on the first Kalman gain to obtain a position state estimation value; in a velocity estimation layer In the method, the first-order difference of the pre-processed relative displacement data is performed to obtain velocity data, the velocity data is input as a state vector, and a state prediction is performed through a state prediction module of the velocity estimation layer to obtain a velocity prediction value; in the second measurement update module of the velocity estimation layer, a second Kalman gain is calculated based on the velocity prediction value and the velocity observation value; in the second state fusion module of the velocity estimation layer, the velocity prediction value is corrected based on the second Kalman gain to obtain a velocity state estimation value; in the acceleration estimation layer, the first-order difference of the velocity data is performed to obtain acceleration data, the acceleration data is input as a state vector, and a state prediction is performed through a state prediction module of the acceleration estimation layer to obtain an acceleration prediction value; in the third measurement update module of the acceleration estimation layer, a third Kalman gain is calculated based on the acceleration prediction value and the acceleration observation value; in the third state fusion module of the acceleration estimation layer, the acceleration prediction value is corrected based on the third Kalman gain to obtain an acceleration state estimation value; the position state estimation value, the velocity state estimation value and the acceleration state estimation value are weightedly fused to obtain terminal position data; The positioning module is used to input the terminal location data, the geomagnetic field sensor data and the wireless signal receiver data into a multi-source data fusion model for three-dimensional spatial positioning to obtain the three-dimensional spatial positioning result of the explosion-proof intelligent terminal; specifically comprising: performing time synchronization and data alignment on the terminal location data, the geomagnetic field sensor data and the wireless signal receiver data to obtain aligned multi-source data; inputting the aligned multi-source data into the feature extraction layer of the multi-source data fusion model, performing feature extraction through a convolutional neural network to obtain a multi-source feature map, and performing self-attention mechanism analysis on the multi-source feature map to calculate the correlation weights between features to obtain a weighted feature map; input the weighted feature map into the fusion layer of the multi-source data fusion model, interact and fuse the features from different sources through the cross attention mechanism to obtain fusion features, and perform residual connection and layer normalization on the fusion features to obtain optimized fusion features; input the optimized fusion features into the first fully connected sublayer of the three-dimensional space mapping layer, perform nonlinear transformation through the ReLU activation function to obtain the first hidden layer output, and perform batch normalization on the first hidden layer output to obtain a standardized first hidden layer output; input the standardized first hidden layer output into the second fully connected sublayer of the three-dimensional space mapping layer, and perform Leaky The ReLU activation function is used to extract features to obtain the output of the second hidden layer, and the second hidden layer output is subjected to dropout regularization to obtain the regularized second hidden layer output; the regularized second hidden layer output is input into the third fully connected sublayer of the three-dimensional space mapping layer, and feature mapping is performed through the Tanh activation function to obtain the output of the third hidden layer; the third hidden layer output is subjected to residual connection, and the third hidden layer output is added to the optimized fusion feature to obtain the residual enhanced feature representation; the residual enhanced feature representation is input into the output layer of the three-dimensional space mapping layer, and the feature is mapped to the three-dimensional space coordinate system through linear transformation to obtain the initial three-dimensional position coordinates; the initial three-dimensional position coordinates are subjected to spatial constraint optimization through conditional random fields to obtain the optimized three-dimensional position coordinates, and the optimized three-dimensional position coordinates are input into the Kalman smoother to smooth the positions of multiple time steps to obtain the three-dimensional space positioning result of the explosion-proof intelligent terminal.

Images