Weak network point detection method based on course angle and WiFi sequence

Abstract

The invention discloses a weak network point detection method based on a course angle and a WiFi sequence, and the method comprises the steps: constructing a motion track according to a course angle feature sequence generated in the moving process of a user, recording the motion track in a preorder path of a weak network point region and WiFi signal strength evolution data when the user enters the weak network point, and carrying out the detection of the weak network point. Establishing a space-time correlation model between the trajectory model and the signal strength evolution data, constructing a weak network scene feature template library, and when the user moves again, predicting whether the user is about to enter a weak network area by analyzing the similarity between the change trend of the current wireless signal and historical features in the weak network scene feature template library. And a network switching mechanism is started in advance. The joint recognition model is established on the basis of keeping independence of two types of data, specific position information of a user does not need to be obtained, prediction and judgment are achieved by analyzing internal association of moving behaviors and signal changes, and the method can be widely applied to complex building environments such as shopping malls, transportation hubs and underground spaces.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communication, and particularly relates to a method for realizing intelligent prediction and active switching in weak network areas through a multi-stage collaborative processing architecture. Background Art

[0002] With the development of the mobile Internet, WiFi and cellular networks have become the main ways for intelligent terminal users to access the network. In indoor environments, WiFi networks can provide relatively high actual data transmission rates within a short distance and have low deployment costs, so they are widely used. At some locations, the WiFi signal is weak, resulting in a decline in network quality or even connection interruption. These locations with weak signals can be called weak network points. When users move to these areas, if they fail to switch to the cellular network in time, situations such as video stuttering, poor call quality, or data transmission failure may occur, leading to a decline in user experience. Therefore, how to accurately identify weak network points and trigger the switching between WiFi and cellular networks in advance to ensure the network connection quality of users is an important research issue.

[0003] Currently, the switching technology between WiFi and cellular networks mainly relies on the WiFi signal strength and the base station signal quality (such as signal-to-noise ratio, throughput, etc.) for judgment. Common strategies include:

[0004] (1) Network switching strategy based on a threshold: Usually, a fixed WiFi signal strength threshold (such as -80 dBm) is set. When the terminal detects that the WiFi signal is lower than this threshold, it triggers a switch to the cellular network. However, since the WiFi signal is greatly affected by the environment, such as factors like wall obstruction, access point deployment location, and interference sources, the signal strength fluctuates greatly. In indoor environments, even if the user is still within the available WiFi coverage area, the signal strength may briefly drop, resulting in mis-switching. At the same time, the fixed threshold is difficult to adapt to different network environments, and the optimal switching threshold may be different in different scenarios, resulting in a lack of flexibility in the switching strategy.

[0005] (2) Prediction of WiFi signals based on machine learning: Some studies use machine learning methods to analyze the changing trend of WiFi signals and predict whether the user is about to enter a weak network point to optimize the network switching strategy. However, this method relies on historical RSSI data for prediction, and the WiFi signal is severely interfered by the environment. Predicting only based on signal strength data is easily affected by instantaneous signal fluctuations, resulting in misjudgment. At the same time, this method does not fully consider the user's moving direction and path information and cannot accurately identify whether the area the user is about to enter is a weak network point.

[0006] (3)WiFi Fingerprint Location: This method usually relies on building an indoor WiFi fingerprint database, using machine learning or signal matching algorithms to estimate the user's current location, and combining the WiFi signal strength to determine whether the user enters a weak signal area. However, such methods have significant drawbacks. First, in terms of privacy, it is necessary to collect and store the user's location information, and location information belongs to sensitive privacy data, so users are reluctant to provide it. Some solutions process the location data through a cloud server, increasing the risk of data leakage and abuse. Second, the deployment cost is high. WiFi fingerprint location requires pre-collecting a large number of WiFi signal characteristics and establishing an indoor fingerprint database. Since the WiFi environment is easily affected by factors such as device changes and obstacle movements, the fingerprint database needs to be updated frequently, resulting in high maintenance costs and making it difficult to promote on a large scale. Finally, the computational complexity is high. WiFi fingerprint location usually requires complex matching calculations on the terminal or server side, increasing the computational burden and energy consumption. When the user moves, there may be a delay in real-time location, affecting the timeliness of network switching.

[0007] Generally speaking, the existing network switching strategies mainly rely on the WiFi signal strength, without fully utilizing the user's moving direction information and the WiFi signal strength change sequence to comprehensively analyze the location of weak signal areas. This leads to the following main problems:

[0008] (1)Poor privacy protection: It is necessary to collect and upload location information, which brings the risk of privacy leakage.

[0009] (2)High misjudgment rate: The WiFi signal fluctuates complexly, and the traditional switching strategy based on the RSSI threshold is prone to false triggering or delayed triggering of switching. Summary of the Invention

[0010] In view of the above problems existing in the prior art, the present invention provides a method for identifying weak signal areas based on moving direction features and wireless signal strength evolution. By collecting the user's heading angle sequence and WiFi signal strength sequence, analyzing the variation laws of the user's motion state and wireless signals, realizing the intelligent prediction of communication weak signal areas, and automatically switching the network access mode before the user approaches the weak network area to ensure communication continuity.

[0011] To achieve the above object, the present invention provides the following technical solutions:

[0012] A method for detecting weak signal areas based on heading angle and WiFi sequence, comprising the following steps:

[0013] Step 1: Establish a multi-modal data acquisition system on the mobile terminal, capture the heading angle feature sequence in real time through the inertial navigation module, and collect the WiFi signal strength sequence by using the wireless communication module;

[0014] Step 2: When the user enters a weak signal area, record the heading angle sequence and the WiFi signal strength sequence in the previous path of this area, and establish a spatio-temporal correlation model between the movement trajectory and the evolution data of the WiFi signal strength;

[0015] Step 3: Use the sequence matching algorithm and pattern classification technology to identify the characteristic rules of the weak signal area, construct the feature space of the stored weak signal area data set, and build a weak network scenario feature template library;

[0016] Step 4: When the user moves again, perform multi-dimensional comparison between the real-time movement trajectory and the weak network scenario feature template library. When it is detected that the user's trajectory matches the previous trajectory of the known weak signal area, by analyzing the similarity between the current change trend of the WiFi signal and the historical features in the weak network scenario feature template library, predict whether the user is about to enter a weak signal area.

[0017] Further, in Step 1, a movement trajectory model based on the heading angle feature sequence is constructed by a composite algorithm that fuses gyroscope angular velocity integration, accelerometer movement direction detection, and magnetometer azimuth correction. The movement trajectory model uses spatio-temporal features including heading angle change rate, path curvature, and movement vector angle to establish a trajectory fingerprint library without geographical coordinates.

[0018] Further, in Step 1,

[0019] The heading angle feature sequence is expressed as:

[0020] ;

[0021] where is the weight factor, is the gyroscope integral value, is the magnetometer estimated value;

[0022] For the WiFi signal strength sequence, an event-triggered sampling strategy based on motion state detection or a hierarchical sampling strategy based on speed threshold is adopted to dynamically adjust the scanning frequency.

[0023] Further, in Step 2, independent feature analysis is performed on the heading angle sequence and the WiFi signal strength sequence respectively. The former extracts kinematic parameters including heading angle change features and vector trajectory similarity, and the latter analyzes signal attenuation gradient, fluctuation periodicity, and time domain differential features; the correlation of the sequences is analyzed by the dynamic time warping algorithm to establish a cross-modal and position-independent spatio-temporal correlation model.

[0024] Furthermore, in step 3, a motion trajectory sequence based on the heading angle sequence is constructed, and a WiFi signal evolution feature sequence based on time-frequency domain characteristics and frequency domain characteristics is constructed. The dynamic time warping algorithm is used to establish the correlation matrices of the heading angle sequence, the motion trajectory sequence, the WiFi signal strength sequence and the WiFi signal evolution feature sequence respectively; combined with the correlation matrix, the hierarchical clustering method is used to perform spatial division of the high-dimensional feature vector, automatically identify the weak network area categories with similar building structure characteristics, and generate a feature template library.

[0025] Furthermore, in step 4, a multi-stage recognition strategy is adopted. In the first stage, the current heading angle feature sequence and the feature library are subjected to sliding window similarity calculation, and the dynamic time warping algorithm is used to evaluate the trajectory matching degree and screen potential weak network areas. In the second stage, the dynamic time warping algorithm is used to calculate the correlation between the current signal sequence and the signal sequence of the feature template library, and the signal trend similarity is calculated for secondary verification. A confidence transfer mechanism is set between the two stages, and an early warning is triggered only when both the trajectory matching degree and the signal similarity exceed the correlation threshold.

[0026] Furthermore, in step 4, when the user's travel trajectory and signal characteristics both meet the characteristics of a weak network point, the network switching strategy is initiated in advance, switching to the cellular network before the signal quality decreases, and switching back to WiFi after the signal is restored.

[0027] Furthermore, it also includes a dynamic update step of the weak network point database, adopts a sliding window and a long-term data analysis mechanism to regularly update the weak network point database to adapt to the dynamic changes of the WiFi environment, and improves the accuracy of the recognition model through the continuous accumulation of user data.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The method is systematically verified through empirical research on typical indoor scenes (including complex electromagnetic environments such as high-frequency occlusion areas and areas with dense metal structures). In test areas with significant building structure characteristics (such as semi-enclosed spaces with prominent multipath effects), the system demonstrates excellent scene recognition capabilities, with weak point recognition accuracy and recall rates both exceeding 95%. The terminal storage occupancy is less than 10% of that of mainstream positioning solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of a weak network point detection method based on heading angle and WiFi sequence of the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0032] The core idea of the present invention is as follows: First, a motion trajectory is constructed based on the heading angle feature sequence generated during the user's movement. When the user enters a weak signal area, the motion trajectory and the evolution data of the WiFi signal strength in the previous path of this area are recorded, a spatio-temporal correlation model between the trajectory model and the signal strength evolution data is established, and the sequence matching algorithm and pattern classification technology are used to identify the characteristic rules of the weak signal area, and a feature template library of typical weak signal scenarios is constructed. When the user moves again, the system performs multi-dimensional comparison between the real-time motion trajectory and the weak signal scenario feature library. When it is detected that the user's trajectory matches the previous trajectory of the known weak signal area, by analyzing the similarity between the current change trend of the wireless signal and the historical features in the weak signal scenario feature library, it is predicted whether the user is about to enter a weak signal area, and the network switching mechanism is activated in advance.

[0033] The present invention adopts a multi-stage recognition strategy, decouples the trajectory feature analysis and the signal evolution analysis in time sequence, and establishes a joint recognition model on the basis of maintaining the independence of the two types of data. This solution does not require obtaining the specific location information of the user, realizes prediction and judgment by analyzing the internal relationship between the movement behavior and the signal change, effectively protects the user privacy while reducing the computing load of the device. This technology can be widely applied to complex building environments such as shopping malls, transportation hubs, and underground spaces, providing an adaptive network optimization solution for mobile terminals and significantly improving the communication experience of users in mobile scenarios.

[0034] The method of the present invention realizes the intelligent prediction and active switching of weak signal areas through a multi-stage collaborative processing architecture, including the following core execution processes (as Figure 1 shown):

[0035] Step 1: Multi-modal data collection

[0036] A multi-modal data collection system is established on the mobile terminal. The heading angle sequence is captured in real time through an inertial navigation unit, and the WiFi signal strength sequence is collected by using a wireless communication module. Among them, the heading angle sequence is generated by a composite algorithm that fuses the gyroscope angular velocity integration, the accelerometer motion direction detection, and the magnetometer azimuth correction to ensure the spatio-temporal continuity of the motion vector. According to the three-axis data of the accelerometer , the attitude angle of the device is estimated, including (pitch angle and roll angle ):

[0037]

[0038]

[0039] Assume the original magnetometer data is , then the magnetometer is compensated according to the attitude angle to obtain the magnetic component on the horizontal plane:

[0040]

[0041]

[0042] Using the compensated magnetometer components ( , calculate the heading angle :

[0043]

[0044]

[0045] where is the normalized heading angle, in degrees.

[0046] The angular velocity measured by the gyroscope can be used to perform short-term integral prediction on :

[0047]

[0048] To avoid the cumulative error of the gyroscope, the present invention adopts a complementary filtering method for heading angle fusion and correction:

[0049]

[0050] where is the weight factor, is the gyroscope integral value, is the magnetometer estimated value.

[0051] The signal strength sequence adopts an event-triggered sampling strategy based on motion state detection or a hierarchical sampling strategy based on speed threshold. The event-triggered sampling strategy based on motion state detection dynamically adjusts the scanning frequency according to the terminal moving speed. By adopting the event-triggered sampling strategy based on motion state detection, combining the accelerometer and gyroscope data to detect in real time whether the user is in a stationary, walking, running, or riding state. Dynamically enable different sampling frequencies according to the identified motion state, low-frequency or paused sampling in the stationary state; high-frequency scanning when moving fast; trigger a sampling immediately when the state changes. The hierarchical sampling strategy based on speed threshold derives the current moving speed through the accelerometer or position estimation, sets multiple speed thresholds, and when the speed is in different intervals, adopts different scanning frequencies. This stage outputs a trajectory-signal multimodal data stream, providing the original input for subsequent analysis.

[0052] Step 2: Weak network scenario triggered feature storage

[0053] When it is detected that the user manually switches the WiFi signal to the cellular signal, the current location is marked as a weak signal point, and the system automatically retraces and stores the heading angle sequence and the corresponding WiFi signal strength sequence within a set time window before the switching trigger. The storage structure adopts a hierarchical index design, including the WiFi access point ID, timestamp, heading angle sequence, and WiFi signal strength sequence, supporting efficient feature retrieval and comparison. This step provides data support for subsequent feature engineering.

[0054] Step 3: Multi-dimensional Feature Engineering

[0055] Construct a feature space for the stored weak signal point dataset to provide a reference feature template library for subsequent online prediction and decision-making. The content includes:

[0056] (1) Extraction of motion trajectory features: Based on the heading angle sequence and combined with acceleration, construct a motion trajectory sequence.

[0057] (2) Analysis of signal evolution features: Extract time-frequency domain features such as the first derivative (attenuation rate), second derivative (change rate of attenuation rate), and power spectral density of the signal strength sequence to construct a feature sequence.

[0058] (3) Correlation analysis: Use the dynamic time warping algorithm to establish the correlation matrices of the heading angle sequence, motion trajectory sequence, signal strength sequence, and the feature sequence of the signal strength sequence respectively.

[0059] Let the set of weak signal points be , and each point corresponds to a heading angle time series:

[0060]

[0061] Use dynamic time warping (DTW) to calculate the shortest alignment distance between any two sequences . First, construct an Euclidean distance matrix , the size of the matrix is , and the matrix element can be expressed as:

[0062]

[0063] Then construct an accumulated distance matrix , the size of the matrix is , and the matrix element can be expressed as:

[0064]

[0065] The final correlation calculation is:

[0066]

[0067] Finally, the correlation matrix of the heading angle channel is obtained:

[0068]

[0069] Perform the same correlation matrix calculation process on the motion trajectory sequence.

[0070] Similarly, the weak network points have a signal strength time series:

[0071]

[0072] Calculate the signal strength sequence the DTW distance between , and construct a signal strength correlation matrix:

[0073]

[0074] Perform the same correlation analysis on the feature sequence of the signal strength sequence.

[0075] (4) Feature space clustering: Combining the correlation matrix, use the hierarchical clustering method to partition the high-dimensional feature vectors in space, automatically identify the weak network area categories with similar building structure characteristics, and generate a feature template library.

[0076] Step 4: Phased prediction and decision fusion

[0077] During the real-time operation phase, the system performs:

[0078] (1) Motion trajectory matching: Calculate the similarity of the current heading sequence and the feature library using a sliding window, and use the dynamic time warping algorithm to evaluate the trajectory matching degree to screen potential weak network areas. The system continuously collects the heading angle sequence of the current time period , and perform dynamic time warping (DTW) calculation with the pre-stored weak network point trajectory sequences in the feature template library in a sliding window manner to obtain the trajectory matching distance , and estimate the trajectory matching degree accordingly:

[0079]

[0080] Among them, is the normalization coefficient to ensure that the similarity value , the higher the matching degree, the more consistent the path, and the higher the potential risk.

[0081] (2) Signal evolution trend analysis: Calculate the correlation between the current signal sequence and the signal sequences in the feature template library using the dynamic time warping algorithm, and calculate the signal trend similarity. Record the current WiFi signal strength sequence in real time. , and also compare it with the signal evolution curves in the template library through the DTW method to calculate the signal similarity:

[0082]

[0083] Among them, is the minimum path distance obtained by DTW, is the adjustment factor used to control the sensitivity of the signal similarity.

[0084] (3) Risk probability assessment: Establish a Bayesian decision model, fuse the trajectory matching degree, signal trend similarity, and environmental context information, and calculate the probability estimate of entering the weak network area. When the comprehensive risk value exceeds the dynamically adjusted decision threshold, trigger a pre-switch warning.

[0085] Considering the trajectory matching degree , signal trend similarity , and optional environmental context variables (such as time period, indoor / outdoor label, historical network fluctuation frequency, etc.), construct a joint probability model to predict the probability of currently entering the weak network area , and the conditional probability formula is as follows:

[0086]

[0087] represents the current weak network area, represents the non-weak network area; is the normalization factor (i.e., the denominator of the total probability formula); , are probability density estimation models constructed through historical training samples; is the prior probability of the occurrence of a weak network, which can be dynamically updated through historical frequencies; is the environmental context conditional probability. Finally, the system records the estimated probability of entering the weak network area calculated as:

[0088]

[0089] When , that is, higher than the dynamically adjusted risk threshold , the system triggers the warning mechanism and recommends implementing the network switching preparation strategy.

[0090] The trajectory matching degrees of all trajectories marked as weak network points Sum signal similarity Construct a sample set, and use the kernel density estimation method to model the similarity probability to obtain a continuous probability density function. Taking as an example:

[0091]

[0092] Among them, is the th sample (weak network trajectory matching degree), is the bandwidth parameter (which can be selected by the Silverman rule or cross-validation), is the kernel function, and the commonly used one is the Gaussian kernel .

[0093] Process in the same way, as well as the under non-weak network samples .

[0094] Environmental context conditional probability. Typical context variables include time period (such as morning, noon, evening, weekday / weekend), user activity pattern (stationary / walking / cycling), which is expressed as:

[0095]

[0096] Step 5: Online incremental learning and model evolution

[0097] Build a closed-loop optimization system to achieve the dynamic evolution of the system's prediction ability, including:

[0098] (1) Concept drift detection: Monitor the deviation between historical prediction results and the actual network state, and identify the feature drift phenomenon caused by environmental transformation.

[0099] (2) Incremental feature learning: Perform online feature extraction on the new weak network event data stream, and achieve the dynamic evolution of the template library through template weight adjustment.

[0100] (3) Decaying template elimination: Adopt a sliding window evaluation mechanism to calculate the confidence decay of the feature templates that have not been triggered for a long time, and automatically clean up the invalid templates.

[0101] To sum up, in view of the privacy protection defect caused by the traditional weak network identification technology relying on geographical coordinate positioning, the present invention creatively proposes a location-free identification mechanism based on mobile direction features. By continuously collecting the heading angle parameters through the azimuth sensor built in the intelligent terminal, a motion trajectory model with spatio-temporal continuity is constructed. This model depicts the user's movement law through features such as the rate of change of the direction angle and the included angle of the motion vector, completely avoiding the collection and transmission of geographical coordinate information.

[0102] The terminal device obtains the heading angle of the user during movement through a built-in orientation sensor , and combines it with a fixed time interval for sampling to obtain a sequence:

[0103]

[0104] Based on the collected changes in the heading angle and the step length estimated by the user assuming uniform speed or combined with the step frequency , a relative displacement trajectory in the two-dimensional plane is constructed . The initial position is set as , then:

[0105]

[0106]

[0107] where can be estimated from the acceleration data

[0108] The above point set constitutes the user's movement trajectory path:

[0109]

[0110] Combined with the timestamp , it is extended to a spatio-temporal trajectory:

[0111]

[0112] At the same time, time-domain and frequency-domain features are extracted from the WiFi signal strength sequence. The features of the signal attenuation pattern are extracted in the time domain dimension, and the periodic features of the signal fluctuation are analyzed in the frequency domain dimension. Correlation analysis is performed on the movement trajectory sequence and the signal strength sequence respectively to establish a weak network area recognition model without position dependence, realizing accurate environmental perception under the premise of protecting user privacy in complex indoor scenarios such as shopping malls and underground passages

[0113] In addition, in view of the problem of scene confusion caused by simply relying on signal strength threshold judgment in the prior art, the present invention has developed a motion-signal multi-modal dynamic association technology. By establishing a cross-modal association model between the motion trajectory sequence and the signal strength evolution sequence, the coupling law between the direction offset vector and the signal gradient change is revealed: when the user approaches a complex building structure area, the mutation characteristics of the direction vector will generate a temporal correlation characteristic with the non-linear attenuation of the signal strength. Based on the dynamic time warping algorithm, the correlation between the motion trajectory change sequence and the signal strength evolution sequence is analyzed respectively; through the hierarchical clustering method, the spatial distribution of the matched feature vectors is modeled, and the weak network area categories with similar building structure characteristics are automatically identified. This technology realizes the cross-domain feature fusion of motion behavior patterns and signal evolution laws for the first time, enabling the system to effectively distinguish between temporary signal occlusion and persistent weak network areas caused by building structures, and significantly improving the recognition accuracy in dynamic complex scenarios such as airport terminals and multi-level interchange hubs.

[0114] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A weak point detection method based on the heading angle and WiFi sequence, characterized in that, It includes the following steps: Step 1: Establish a multi-modal data acquisition system on the mobile terminal. Real-time capture the heading angle feature sequence through the inertial navigation module, and collect the WiFi signal strength sequence using the wireless communication module; Step 2: When the user enters a weak signal area, record the heading angle sequence and the WiFi signal strength sequence in the previous path of this area, and establish a spatio-temporal correlation model between the motion trajectory and the evolution data of the WiFi signal strength; Step 3: Use the sequence matching algorithm and pattern classification technology to identify the characteristic laws of the weak signal area, construct the feature space of the stored weak signal area data set, and construct a weak signal scenario feature template library; Step 4: When the user moves again, perform multi-dimensional comparison between the real-time motion trajectory and the weak signal scenario feature template library. When it is detected that the user's trajectory matches the previous trajectory of the known weak signal area, by analyzing the similarity between the current change trend of the WiFi signal and the historical features in the weak signal scenario feature template library, predict whether the user is about to enter a weak signal area.

2. The weak point detection method based on heading angle and WiFi sequence according to claim 1, wherein In Step 1, a motion trajectory model based on the heading angle feature sequence is constructed through a composite algorithm that fuses gyroscope angular velocity integration, accelerometer motion direction detection, and magnetometer azimuth correction. The motion trajectory model uses spatio-temporal features including the heading angle change rate, path curvature, and motion vector angle to establish a trajectory fingerprint library without geographical coordinates.

3. The weak point detection method based on the heading angle and the WiFi sequence according to claim 1, wherein In Step 1, The heading angle feature sequence is expressed as: ； Among them, is the weight factor, is the gyroscope integral value, is the magnetometer estimated value; For the WiFi signal strength sequence, an event-triggered sampling strategy based on motion state detection or a hierarchical sampling strategy based on speed threshold is adopted to dynamically adjust the scanning frequency.

4. The weak point detection method based on the heading angle and the WiFi sequence according to claim 1, wherein In Step 2, independent feature analysis is performed on the heading angle sequence and the WiFi signal strength sequence respectively. The former extracts kinematic parameters including heading angle change features and vector trajectory similarity, and the latter analyzes signal attenuation gradient, fluctuation periodicity, and time-domain differential features; Analyze the correlation of the sequences through the dynamic time warping algorithm to establish a cross-modal and position-independent spatio-temporal correlation model.

5. The weak point detection method based on heading angle and WiFi sequence according to claim 1, wherein In Step 3, construct a motion trajectory sequence based on the heading angle sequence, construct a WiFi signal evolution feature sequence based on time-frequency domain features and frequency domain features, and use the dynamic time warping algorithm to establish the correlation matrices of the heading angle sequence, the motion trajectory sequence, the WiFi signal strength sequence, and the WiFi signal evolution feature sequence respectively; Combine the correlation matrices, and use the hierarchical clustering method to divide the high-dimensional feature vectors in space, automatically identify the weak signal area categories with similar building structure characteristics, and generate a feature template library.

6. The weak point detection method based on the heading angle and the WiFi sequence according to claim 1, characterized in that In Step 4, adopt a multi-stage recognition strategy. In the first stage, calculate the sliding window similarity between the current heading angle feature sequence and the feature library, use the dynamic time warping algorithm to evaluate the trajectory matching degree, and screen potential weak signal areas; In the second stage, use the dynamic time warping algorithm to calculate the correlation between the current signal sequence and the signal sequence in the feature template library, and calculate the signal trend similarity for secondary verification; A confidence transfer mechanism is set between the two stages, and an early warning is triggered only when both the trajectory matching degree and the signal similarity exceed the correlation threshold.

7. The weak point detection method based on the heading angle and the WiFi sequence according to claim 1, wherein In step 4, when both the user's movement trajectory and signal characteristics meet the weak network point characteristics, the network switching strategy is started in advance, switched to the cellular network before the signal quality drops, and switched back to WiFi after the signal is restored.

8. The weak point detection method based on heading angle and WiFi sequence according to claim 1, wherein It also includes the dynamic update step of the weak network point database. By using a sliding window and long-term data analysis mechanism, the weak network point database is updated regularly to adapt to the dynamic changes of the WiFi environment. Through the continuous accumulation of user data, the accuracy of the recognition model is improved.

Images