A wind speed monitoring and identification method and system based on pattern matching

Abstract

The present invention relates to the field of wind speed monitoring technology, and discloses a wind speed monitoring and identification method and system based on pattern matching. The method comprises: decomposing a real-time wind speed signal into a steady-state component and a transient component, generating a spatial weight matrix in combination with terrain features to suppress local interference; dynamically allocating matching weights based on signal energy distribution and historical matching confidence; matching the optimal candidate pattern from a steady-state and transient pattern library, calculating the comprehensive matching degree to determine the final result; and performing incremental learning and updating on mismatched transient signal features. The present invention effectively overcomes the problem of wind speed signal distortion under complex terrain through the synergistic effect of spatiotemporal feature decoupling and dynamic weight allocation, enables the pattern matching process to adaptively focus on key features, and significantly improves monitoring accuracy.

Description

Technical Field

[0001] The invention relates to a wind speed monitoring and identification method and system based on pattern matching, belonging to the technical field of wind speed monitoring. Background Art

[0002] In scenarios such as bridge health monitoring and meteorological environmental monitoring, the dynamic feature extraction and pattern matching accuracy of wind speed signals directly impact the reliability of the monitoring system. Existing technologies typically match the steady-state and transient components of wind speed signals using predefined pattern libraries, but this presents significant challenges in complex terrain environments and long-term monitoring scenarios.

[0003] Currently, the industry mainly adopts the following technical paths: first, separate the steady-state and transient components of the wind speed signal through Fourier transform or wavelet decomposition, and then match them with the pre-stored pattern library based on a fixed weight allocation strategy. However, this approach has three limitations: 1. The static pattern library cannot dynamically adapt to signal propagation distortion caused by terrain changes. When the distribution of obstacles around the monitoring point or the wind direction suddenly changes, the matching results are easily affected by local interference; 2. The fixed weight allocation strategy ignores the dynamic correlation between signal energy distribution and historical matching credibility, resulting in an increased mismatch rate in sudden turbulence events; 3. The traditional pattern library update mechanism relies on full data return and offline training, which makes it difficult to achieve real-time incremental learning on embedded devices. Pattern library aging problems are prone to occur in long-term monitoring.

[0004] To address the above issues, existing improvement solutions attempt to introduce dynamic attenuation coefficients to adjust the strength of terrain interference suppression, or to update the pattern library through regular retraining. However, such methods still have new defects: manual calibration of the attenuation coefficient is difficult to adapt to the changing actual terrain conditions, and frequent retraining of the pattern library will lead to the loss of historical valid patterns, while significantly increasing computing resource consumption. For example, in the monitoring scenario of wind farms in mountainous areas, existing methods are unable to integrate terrain projection features with signal energy distribution in real time, and often misjudge transient turbulence caused by flow around the mountain as equipment abnormality signals, resulting in frequent false alarms in the monitoring system. Summary of the Invention

[0005] The present invention provides a wind speed monitoring and identification method and system based on pattern matching, the main purpose of which is to solve the problems of poor adaptability of static pattern library, insufficient dynamic interference suppression capability and long-term monitoring performance attenuation.

[0006] To achieve the above object, the present invention provides a wind speed monitoring and identification method based on pattern matching, comprising the following steps:

[0007] Spatiotemporal feature decoupling step: The acquired real-time wind speed signal is decomposed into a steady-state component representing the long-term trend and a transient component representing the instantaneous fluctuation. Based on the terrain feature information around the wind speed monitoring point, a spatial weight matrix is ​​generated to suppress the impact of local interference on the global pattern.

[0008] Dynamic weight allocation step: based on the signal energy distribution of the steady-state component and the transient component and the historical matching confidence , calculate and distribute the matching weights of the steady-state component and the transient component in the subsequent pattern matching process in real time, wherein the historical matching confidence The number of successful matches within the preset time window based on the current mode Total matches The ratio of is determined, that is , and when the historical matching confidence is lower than a preset threshold, triggering the incremental learning update of the transient pattern library;

[0009] Pattern library matching step: matching the candidate patterns that are most similar to the steady-state component and the transient component from the pre-built steady-state pattern library and the transient pattern library, respectively, and calculating the comprehensive matching degree in combination with the spatial weight matrix to determine the final wind speed monitoring identification result;

[0010] Incremental learning update step: For transient signal features that are continuously mismatched, their feature information is extracted, and the transient pattern library is locally iteratively updated based on the extracted feature information. The local iterative update includes comparing the transient signal features within the mismatch period with the existing transient pattern library for similarity. If the similarity is lower than a preset threshold, it is added as an independent pattern; otherwise, it is merged into the nearest neighbor pattern, and the steady-state pattern library remains unchanged.

[0011] In a preferred embodiment, the method for generating the spatial weight matrix in the spatiotemporal feature decoupling step includes: obtaining the height information and distribution density information of obstacles within a preset distance range around the wind speed monitoring point; Angle from the current wind direction , dynamically adjust the attenuation coefficient of wind speed signals in each direction , the attenuation coefficient The angle There is a positive correlation.

[0012] In a preferred embodiment, in the dynamic weight allocation step, the preset time window is the past 30 seconds, and the preset threshold of the historical matching confidence is 0.5.

[0013] In a preferred embodiment, in the spatiotemporal feature decoupling step, a lightweight Fourier transform is used to decompose the real-time wind speed signal to obtain the steady-state component and the transient component.

[0014] In a preferred embodiment, in the pattern library matching step, the comprehensive matching degree is calculated on the embedded device by using a table lookup method instead of real-time matrix operations to reduce computational complexity.

[0015] In a preferred embodiment, after the real-time wind speed signal is decomposed into a steady-state component and a transient component in the spatiotemporal feature decoupling step, a morphological feature self-screening mechanism is also included: a morphological template library is constructed, and the morphological template library is based on the transient signal waveform of historical turbulence events, and its quantitative features of local extreme value distribution, waveform symmetry and energy attenuation slope are extracted, and it contains no less than five typical turbulence templates; the current transient signal is dynamically time-regularly matched with each template in the morphological template library, and the waveform morphology similarity is calculated; the morphological matching threshold is dynamically adjusted according to the current ambient noise level, wherein the ambient noise level is obtained by inversely inferring the energy entropy of the steady-state component, and only the candidate turbulence templates with a matching degree higher than the morphological matching threshold are retained to enter the subsequent dynamic weight allocation process.

[0016] In a preferred embodiment, the method for constructing the morphological template library further includes a dynamic template self-organizing mechanism, including: when the proportion of transient signal features that cannot match the existing morphological template library in 10 consecutive incremental learning updates exceeds 30%, the reconstruction of the morphological template library is automatically triggered; a density-based online clustering algorithm is used to merge the newly added transient signal features and historical templates into superclasses according to morphological similarity, and to delete outdated templates that have not been matched in the last six months; according to the quarterly change trend of the steady-state component energy entropy, the weight ratio of each type of typical turbulence template in the morphological template library is dynamically adjusted.

[0017] In a preferred embodiment, in the dynamic template self-organizing mechanism, a sliding window density peak detection method is used for online clustering, retaining only the most representative morphological features in the current window, and the retained proportion does not exceed 20% of the peak value, so as to control the memory usage within a preset range.

[0018] In a preferred embodiment, the method for calculating the historical matching confidence is: counting the ratio of the number of successful matches of the current pattern in the past 30 seconds to the total number of matches as the historical matching confidence, and the method is applied to the field of bridge structure health monitoring or meteorological environment monitoring.

[0019] In a preferred embodiment, a wind speed monitoring and identification system based on pattern matching is also included, and the system includes: a spatiotemporal feature decoupling module, configured to decompose the acquired real-time wind speed signal into a steady-state component representing a long-term trend and a transient component representing an instantaneous fluctuation, and generate a spatial weight matrix for suppressing the influence of local interference on the global pattern based on the terrain feature information around the wind speed monitoring point; a dynamic weight allocation module, configured to calculate and allocate the matching weights of the steady-state component and the transient component in the subsequent pattern matching process in real time based on the signal energy distribution of the steady-state component and the transient component and the historical matching confidence, wherein the historical matching confidence is determined according to the ratio of the number of successful matches of the current pattern within a preset time window to the total number of matches, and when the historical matching confidence is lower than a preset threshold, the incremental learning update of the transient pattern library is triggered; a pattern library matching module, configured to match the candidate patterns that are most similar to the steady-state component and the transient component from the pre-built steady-state pattern library and the transient pattern library, respectively, and calculate the comprehensive matching degree in combination with the spatial weight matrix to determine the final wind speed monitoring and identification result; The incremental learning update module is configured to extract characteristic information of transient signal features that are continuously mismatched, and perform local iterative updates on the transient pattern library based on the extracted characteristic information. The local iterative updates include comparing the transient signal features within the mismatch period with the existing transient pattern library for similarity. If the similarity is lower than a preset threshold, it is added as an independent pattern; otherwise, it is merged into the nearest neighbor pattern, and the steady-state pattern library remains unchanged.

[0020] Compared with the problems described in the background technology, the beneficial effects of the present invention are:

[0021] 1. The essential attributes of the wind speed signal are hierarchically stripped away through spatiotemporal feature decoupling. During the decoupling process of the steady-state and transient components, the spatial constraint effect of terrain features on the signal propagation path is naturally integrated, enabling the pattern matching process to automatically correct the distortion of the global wind speed pattern caused by local environmental interference. Combined with the dynamic weight allocation mechanism's real-time response to the signal energy distribution and historical confidence, an adaptive feature attention focus is formed to ensure that the matching process always focuses on the most representative feature combination, thus breaking through the limitations of traditional methods in the separate processing of steady-state and transient signal features in complex terrain scenarios.

[0022] 2. A morphological template library-based self-screening mechanism for morphological features achieves precise initial screening of turbulent events through quantitative feature comparison and noise-adaptive threshold adjustment. This process forms a closed-loop feedback loop with the incremental learning module. When newly added transient signal features trigger the reconstruction of the transient pattern library, the online clustering algorithm automatically captures the laws of morphological evolution, ensuring that the morphological template library not only retains a memory of historical typical turbulence templates but also continuously absorbs new features generated by dynamic environmental changes. This dual-track evolutionary strategy effectively addresses the problem of matching performance degradation caused by cumulative environmental changes during long-term monitoring in traditional static pattern libraries.

[0023] 3. By combining a lightweight signal decomposition algorithm with an embedded lookup table, the complexity of the core computational process is reduced to a level that can be deployed on edge devices while ensuring pattern matching accuracy. The fixed design of the steady-state pattern library complements the dynamic updating of the transient pattern library, avoiding the risk of pattern drift caused by overlearning while maintaining the stability of the algorithm iteration process through the physical constraints of the spatial weight matrix. This architecture achieves a dual improvement in monitoring accuracy and system robustness in scenarios with limited computing resources.

[0024] 4. By cross-scale correlation of the energy entropy of the steady-state component with the morphological characteristics of the transient signal, a dynamic mapping relationship between the ambient noise level and the pattern matching threshold is established. This adaptive mechanism, combining data-driven and physical mechanisms, enables the system to autonomously adjust its feature screening strategy when encountering sudden meteorological disturbances or long-term environmental evolution, avoiding the limitations of traditional fixed threshold designs on adaptability to complex working conditions and significantly improving the generalization performance of the monitoring algorithm in all-weather scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a diagram of the spatiotemporal feature decoupling module of the present invention.

[0026] Figure 2 This is a diagram of the core processing engine module of the present invention.

[0027] Figure 3 This is a timing diagram of the interaction between the system modules of the present invention.

[0028] Figure 4 The figure is a flow chart of the wind speed monitoring and identification method based on pattern matching of the present invention.

[0029] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0030] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0031] The present invention provides a method for wind speed monitoring and identification based on pattern matching, comprising the following steps:

[0032] Spatiotemporal feature decoupling step: The acquired real-time wind speed signal is decomposed into a steady-state component representing the long-term trend and a transient component representing the instantaneous fluctuation. Based on the terrain feature information around the wind speed monitoring point, a spatial weight matrix is ​​generated to suppress the impact of local interference on the global pattern.

[0033] Dynamic weight allocation step: based on the signal energy distribution of the steady-state component and the transient component and the historical matching confidence , calculate and distribute the matching weights of the steady-state component and the transient component in the subsequent pattern matching process in real time, wherein the historical matching confidence The number of successful matches within the preset time window based on the current mode Total matches The ratio of is determined, that is , and when the historical matching confidence is lower than a preset threshold, triggering the incremental learning update of the transient pattern library;

[0034] Pattern library matching step: matching the candidate patterns that are most similar to the steady-state component and the transient component from the pre-built steady-state pattern library and the transient pattern library, respectively, and calculating the comprehensive matching degree in combination with the spatial weight matrix to determine the final wind speed monitoring identification result;

[0035] Incremental learning update step: For transient signal features that are continuously mismatched, their feature information is extracted, and the transient pattern library is locally iteratively updated based on the extracted feature information. The local iterative update includes comparing the transient signal features within the mismatch period with the existing transient pattern library for similarity. If the similarity is lower than a preset threshold, it is added as an independent pattern; otherwise, it is merged into the nearest neighbor pattern, and the steady-state pattern library remains unchanged.

[0036] In a preferred embodiment, the method for generating the spatial weight matrix in the spatiotemporal feature decoupling step includes: obtaining the height information and distribution density information of obstacles within a preset distance range around the wind speed monitoring point; Angle from the current wind direction , dynamically adjust the attenuation coefficient of wind speed signals in each direction , the attenuation coefficient The angle There is a positive correlation.

[0037] In a preferred embodiment, in the dynamic weight allocation step, the preset time window is the past 30 seconds, and the preset threshold of the historical matching confidence is 0.5.

[0038] In a preferred embodiment, in the spatiotemporal feature decoupling step, a lightweight Fourier transform is used to decompose the real-time wind speed signal to obtain the steady-state component and the transient component.

[0039] In a preferred embodiment, in the pattern library matching step, the comprehensive matching degree is calculated on the embedded device by using a table lookup method instead of real-time matrix operations to reduce computational complexity.

[0040] In a preferred embodiment, after the real-time wind speed signal is decomposed into a steady-state component and a transient component in the spatiotemporal feature decoupling step, a morphological feature self-screening mechanism is also included, including: constructing a morphological template library, wherein the morphological template library is based on the transient signal waveform of historical turbulence events, extracting its quantitative features of local extreme value distribution, waveform symmetry and energy attenuation slope, and containing no less than five typical turbulence templates; dynamically time-regularly matching the current transient signal with each template in the morphological template library, and calculating the waveform morphological similarity; dynamically adjusting the morphological matching threshold according to the current ambient noise level, wherein the ambient noise level is obtained by inversely inferring the energy entropy of the steady-state component, and only retaining the candidate turbulence templates with a matching degree higher than the morphological matching threshold to enter the subsequent dynamic weight allocation process.

[0041] In a preferred embodiment, the method for constructing the morphological template library further includes a dynamic template self-organizing mechanism, including: when the proportion of transient signal features that cannot match the existing morphological template library in 10 consecutive incremental learning updates exceeds 30%, the reconstruction of the morphological template library is automatically triggered; a density-based online clustering algorithm is used to merge the newly added transient signal features and historical templates into superclasses according to morphological similarity, and to delete outdated templates that have not been matched in the last six months; according to the quarterly change trend of the steady-state component energy entropy, the weight ratio of each type of typical turbulence template in the morphological template library is dynamically adjusted.

[0042] In a preferred embodiment, in the dynamic template self-organizing mechanism, a sliding window density peak detection method is used for online clustering, retaining only the most representative morphological features in the current window, and the retained proportion does not exceed 20% of the peak value, so as to control the memory usage within a preset range.

[0043] In a preferred embodiment, the method for calculating the historical matching confidence is: counting the ratio of the number of successful matches of the current pattern in the past 30 seconds to the total number of matches as the historical matching confidence, and the method is applied to the field of bridge structure health monitoring or meteorological environment monitoring.

[0044] In a preferred embodiment, a wind speed monitoring and identification system based on pattern matching is also included, and the system includes: a spatiotemporal feature decoupling module, configured to decompose the acquired real-time wind speed signal into a steady-state component representing a long-term trend and a transient component representing an instantaneous fluctuation, and generate a spatial weight matrix for suppressing the influence of local interference on the global pattern based on the terrain feature information around the wind speed monitoring point; a dynamic weight allocation module, configured to calculate and allocate the matching weights of the steady-state component and the transient component in the subsequent pattern matching process in real time based on the signal energy distribution of the steady-state component and the transient component and the historical matching confidence, wherein the historical matching confidence is determined according to the ratio of the number of successful matches of the current pattern within a preset time window to the total number of matches, and when the historical matching confidence is When it is lower than a preset threshold, the incremental learning update of the transient pattern library is triggered; the pattern library matching module is configured to match the candidate patterns that are most similar to the steady-state component and the transient component from the pre-built steady-state pattern library and the transient pattern library respectively, and calculate the comprehensive matching degree in combination with the spatial weight matrix to determine the final wind speed monitoring and identification result; the incremental learning update module is configured to extract the characteristic information of the transient signal features of continuous mismatches, and perform local iterative update on the transient pattern library based on the extracted characteristic information. The local iterative update includes comparing the transient signal features within the mismatch period with the existing transient pattern library for similarity. If the similarity is lower than the preset threshold, it is added as an independent pattern, otherwise it is merged into the nearest neighbor pattern, and the steady-state pattern library remains unchanged.

[0045] Example 1: This example specifically describes the specific implementation of the wind speed monitoring and identification method and system based on pattern matching. First, the real-time wind speed signal is decomposed by lightweight Fourier transform, and the signal is divided into steady-state component and transient component. For the steady-state component, it mainly reflects the long-term wind speed trend; while the transient component represents wind speed fluctuations or turbulent events. In order to further improve the accuracy of signal processing, for example, a spatial weight matrix is ​​generated by combining the terrain features around the wind speed monitoring point. The method for generating the spatial weight matrix is ​​to dynamically adjust it according to the distribution of obstacles around the wind speed monitoring point and the influence of wind direction changes on the signal propagation path. For example, for mountainous areas, considering the influence of mountains on wind speed signals, the generated spatial weight matrix can be based on the height information of the obstacles and the angle of wind direction. Dynamic adjustment of attenuation coefficient , in order to suppress the influence of local interference caused by obstacle occlusion on the global pattern.

[0046] In the formula, the attenuation coefficient Angle There is a positive correlation between them, namely: ,in, The attenuation function is established based on the actual terrain characteristics. By dynamically calculating and adjusting the wind speed signal attenuation coefficient in each direction, the accuracy of signal decomposition is ensured.

[0047] For the matching of steady-state components and transient components, the matching weight can be dynamically adjusted based on the energy distribution of the signal and the historical matching confidence. It is the number of successful matches within a preset time window based on a certain pattern Total matches Calculated by the ratio of:

[0048] ,

[0049] When history matches confidence When the confidence level falls below a preset threshold, an incremental learning mechanism is triggered to update the transient pattern library. Historical matching confidence helps assess the reliability of pattern matching and dynamically adjusts the weights of each component to avoid mismatches. During the pattern library matching process, the steady-state component and the transient component each search for the most similar candidate pattern from their respective pattern libraries and calculate the overall matching degree using a spatial weight matrix. The overall matching degree is calculated using a weighted average method, ensuring the accuracy of wind speed monitoring results. For mismatched transient signal features, for example, the transient pattern library is updated through an incremental learning mechanism. The incremental learning process involves feature extraction of consecutive mismatched transient signals and comparison of the extracted feature information with the existing transient pattern library. If the similarity falls below a preset threshold, the pattern is added as an independent pattern. If the similarity is high, the pattern is merged into the nearest neighbor pattern. This incremental learning mechanism can effectively respond to new environmental features that emerge during long-term monitoring and automatically optimize the transient pattern library, thereby avoiding pattern library aging.

[0050] In addition, in order to further improve the accuracy of turbulence event identification, this embodiment introduces a morphological template library and its morphological feature self-screening mechanism. The morphological template library includes transient signal waveforms based on historical turbulence events, including five typical turbulence templates. During the matching process, the current transient signal is compared with the templates in the morphological template library through the dynamic time warping method, the waveform morphology similarity is calculated, and the matching threshold is adjusted according to the environmental noise level. The morphological feature self-screening mechanism of the morphological template library further enhances the adaptability of the system. When incremental learning triggers the reconstruction of the transient pattern library, the density-based online clustering algorithm is used to automatically capture the morphological evolution law and eliminate obsolete templates to ensure that the morphological template library always maintains memory of historical typical turbulence templates.

[0051] To accommodate the resource constraints of embedded devices, the pattern matching process uses a table lookup method instead of real-time matrix operations to calculate the overall matching degree. This optimization method significantly reduces computational complexity, ensuring the system can operate efficiently in environments with limited computing resources. Furthermore, the fixed design of the steady-state pattern library complements the dynamic updating of the transient pattern library, ensuring high accuracy and robustness of the system. These are all extended implementations known to those skilled in the art.

[0052] Example 2: This example will be combined with the attached Figure 1 To the attached Figure 4 , the technical solution of the present invention is further described in detail. Figure 1 , is the internal structure of the spatiotemporal feature decoupling module. This module receives the input real-time wind speed signal. First, the module decomposes the input real-time wind speed signal and uses a lightweight Fourier transform to decompose the signal into transient and steady-state components. At the same time, the module also includes a terrain impact processing part for analyzing terrain data. The terrain data enters this part in the form of terrain feature information. After obstacle analysis, the height and distribution density information of the obstacles are obtained. Subsequently, the attenuation coefficient is adjusted. This is based on the angle between the projected area of ​​the obstacle and the current wind direction, and the attenuation coefficient of the wind speed signal in each direction is dynamically adjusted. Finally, the information obtained from the above terrain analysis is integrated through the generation of a spatial weight matrix to generate a spatial weight matrix. Through the processing of the spatiotemporal feature decoupling module, the effective separation of the spatiotemporal features in the input signal and the consideration of the terrain impact are achieved.

[0053] Figure 2 The overall architecture of the core processing engine module is presented. The core processing engine module receives raw wind speed signal input from an external source and passes it to the spatiotemporal feature decoupling module for processing. The output of the spatiotemporal feature decoupling module includes a spatial weight matrix, as well as steady-state and transient components. The spatial weight matrix is ​​sent to the pattern library matching module, while the steady-state and transient components are sent to the dynamic weight allocation module, which calculates and outputs a weighted feature vector. The pattern library matching module receives the weighted feature vector and performs steady-state and transient pattern queries. This module also interacts with the steady-state and transient pattern libraries. The output of the pattern library matching module includes an overall matching degree and mismatch feature feedback. The overall matching degree is output as a monitoring result, while the mismatch feature feedback is sent to the incremental learning update module for updating the transient pattern library. The core processing engine module implements wind speed signal processing, feature extraction, pattern matching, and dynamic updating of the transient pattern library.

[0054] Figure 3This is a timing diagram of the system module interactions, illustrating the interaction flow between the various modules in the system. The entire interaction process begins with the signal source, which sends a real-time wind speed signal to the spatiotemporal feature decoupling module. After receiving the signal, the spatiotemporal feature decoupling module decomposes the signal into steady-state and transient components and passes these components to the dynamic weight allocation module. The dynamic weight allocation module sends the component / weight information to the pattern library matching module. After the pattern library matching module completes the matching process, it outputs the monitoring results to the result output module. If a mismatch occurs during the matching process, the mismatch feature is sent to the incremental learning update module, which sends an instruction to the spatiotemporal feature decoupling module to update the transient pattern library. This timing diagram clearly depicts the flow and processing of data between the various modules.

[0055] Figure 4 The overall process of the method of the present invention begins with an initialization step, after which the system receives a wind speed signal. After receiving the signal, signal decoupling is performed. After decoupling is complete, weight assignment is performed. Once weight assignment is complete, the pattern matching step begins. If the pattern matching step succeeds, the recognition result is output, and the process ends. If the match fails, the incremental learning update step begins. This step updates the transient pattern library. After the transient pattern library is updated, pattern matching is repeated until a successful match is found and a recognition result is output.

[0056] Example 3: In this embodiment, the temporal and spatial characteristics of the real-time wind speed signal are decoupled. The wind speed signal is decomposed into a steady-state component and a transient component by lightweight Fourier transform. For the steady-state component, for example, long-term trend information is extracted; for the transient component, short-term fluctuation characteristics are extracted. In order to cope with the influence of complex terrain on the signal, in the decoupling process, for example, a spatial weight matrix is ​​generated by combining terrain information. The matrix is ​​dynamically adjusted by calculating the relationship between the height, distribution density and wind direction angle of the obstacles around the wind speed monitoring point, and the generated attenuation coefficient is related to the angle There is a positive correlation between them, and the specific formula is: ,in, is the attenuation coefficient of the wind speed signal, is the angle between the obstacle projection and the wind direction, This is an attenuation function established based on terrain characteristics. This attenuation function is dynamically adjusted according to actual environmental conditions to effectively suppress signal interference caused by local obstacles, thereby achieving accurate decoupling of the wind speed signal.

[0057] After signal decoupling, a dynamic weight allocation mechanism is used to match the steady-state component and the transient component. The weight allocation is adjusted based on the energy distribution of the signal and the confidence level of the historical matching. The calculation method is:

[0058] ,

[0059] in, Indicates the number of successful matches within the preset time window. is the total number of matches. When it is lower than the preset threshold (for example, 0.5), the incremental learning mechanism is triggered to update the transient pattern library. By dynamically adjusting the weights, for example, it is ensured that during the matching process, the model can automatically focus on the most representative features based on historical performance and the characteristics of the current signal. In the pattern library matching step, for example, from the steady-state pattern library and the transient pattern library, a matching algorithm is used to find the candidate pattern that is most similar to the decoupled steady-state component and transient component respectively. The comprehensive matching degree of each pattern library is calculated using the spatial weight matrix. The specific matching process is obtained by the weighted average method: ,in, For comprehensive matching, is the steady-state component matching degree, is the transient component matching degree, and are the weight coefficients for steady-state and transient patterns, respectively. For mismatched transient signal features, the system performs incremental learning based on a similarity threshold. Specifically, if the similarity of the mismatched signal falls below a preset value, the signal is added to the transient pattern library as a new independent pattern; otherwise, it is merged into the nearest neighbor pattern. This incremental learning mechanism ensures the dynamic updating of the pattern library, enabling the system to continuously adapt to environmental changes.

[0060] During the transient signal feature matching process, for example, a morphological template library is introduced. Based on the transient signal waveforms of historical turbulence events, a morphological template library containing five typical turbulence templates is constructed. Whenever the system encounters a new turbulence feature, it compares it with the templates in the morphological template library using dynamic time warping to calculate the waveform morphological similarity. To address environmental noise interference, the matching threshold is dynamically adjusted based on the current noise level. When incremental learning is triggered, the dynamic template self-organization mechanism of the morphological template library automatically captures morphological evolution patterns based on an online clustering algorithm. If the proportion of new features that cannot be matched with the existing morphological template library exceeds 30% in 10 consecutive incremental learning cycles, the morphological template library is reconstructed, merging the new templates with historical templates to optimize the library's expressive power. Through this mechanism, the morphological template library can both retain a memory of historical typical turbulence templates and adapt to new environmental changes. These are all extended implementations known to those skilled in the art.

[0061] Example 4: This example further illustrates the incremental learning and transient pattern library update process in a pattern matching-based wind speed monitoring and identification method and system. First, the real-time wind speed signal is decomposed into a steady-state component and a transient component using a lightweight Fourier transform. The steady-state component primarily reflects the long-term trend of wind speed, while the transient component exhibits short-term fluctuations and turbulence characteristics. To address the impact of complex terrain on wind speed signals, real-time terrain information (including obstacle height, distribution density, and wind direction angle) is used to generate a spatial weight matrix. This spatial weight matrix generation step is based on terrain data surrounding the wind speed monitoring point and includes dynamic modeling of the projected area of ​​obstacles and their relationship to wind direction angle to effectively suppress local interference caused by terrain. By introducing this method, signal decomposition not only considers the frequency characteristics of the signal but also incorporates dynamic adjustment of terrain factors. Specifically, the generation of the spatial weight matrix not only relies on static terrain characteristics but also dynamically optimizes the interaction between the wind speed signal and obstacles, thereby providing more accurate signal decoupling results in complex environments.

[0062] After signal decoupling, the steady-state component and the transient component will enter the dynamic weight allocation step respectively. In this step, the matching weight is adjusted in real time based on the energy distribution of the signal and the historical matching confidence. The calculation formula of the historical matching confidence is the ratio of the number of successful matches of the current pattern to the total number of matches. With each match, the historical matching confidence changes dynamically. When the historical matching confidence is lower than a certain threshold, the incremental learning process will be triggered and the transient pattern library will be updated. Through this mechanism, the system can adaptively adjust the matching priority of each signal component, so that the matching process can focus on the most representative wind speed pattern, thereby significantly reducing the risk of mismatching. This dynamic adjustment process not only relies on the past matching success rate, but also combines the characteristics of the real-time signal to make the matching more accurate.

[0063] In order to effectively respond to new features brought about by environmental changes during long-term monitoring, an incremental learning mechanism is introduced into the updating process of the transient pattern library. For transient signals with continuous mismatches, the system extracts their feature information and compares the similarity with the existing transient pattern library. When the similarity is lower than the preset threshold, the newly mismatched signal is added to the transient pattern library as an independent pattern, or it is merged with the nearest neighbor pattern. The key to this incremental learning mechanism is to ensure the dynamic optimization of the transient pattern library through local updates, while avoiding the waste of computing resources caused by frequent full updates. In addition, during the incremental learning process, the system periodically evaluates the newly added patterns and, combined with historical signal characteristics, intelligently adjusts the weights of various patterns in the transient pattern library. Especially in long-term monitoring environments, incremental learning can dynamically introduce new pattern features without affecting the stability of the overall system, effectively avoiding the problem of pattern library aging.

[0064] To improve the accuracy of turbulent event recognition, this embodiment further introduces a morphological template library. Each template in the morphological template library is based on the transient signal waveform of historical turbulent events and extracts features such as local extreme value distribution, waveform symmetry, and energy decay slope. This method allows the system to combine morphological features for initial screening when identifying turbulent signals, reducing noise interference in the recognition process. The dynamic adjustment mechanism of the morphological template library enables it to adapt to new environmental characteristics. When continuous incremental learning fails to match templates in the existing morphological template library, the system analyzes the newly added features using a density-based online clustering algorithm and merges the new features with existing templates based on morphological similarity. This ensures that the morphological template library always retains a memory of historical typical turbulent templates while being able to absorb new characteristic patterns. This mechanism greatly improves the system's adaptability to sudden environmental changes. Furthermore, to ensure efficient operation of the system on embedded devices, this embodiment further optimizes the pattern matching process. Specifically, the system uses a table lookup method instead of real-time matrix operations to calculate the comprehensive matching degree, significantly reducing computational complexity. Furthermore, the system incorporates a lightweight signal decomposition algorithm and embedded computing architecture, enabling high-precision wind speed monitoring and turbulence identification even in environments with limited computing resources. This optimization makes the wind speed monitoring and identification method proposed by the present invention promising for applications in fields such as bridge health monitoring. It ensures real-time performance and stability, particularly in complex terrain or with limited embedded devices. These are all extended implementations known to those skilled in the art.

[0065] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A wind speed monitoring and identification method based on pattern matching, characterized in that: The following steps are involved: Spatiotemporal feature decoupling step: The acquired real-time wind speed signal is decomposed into a steady-state component representing the long-term trend and a transient component representing the instantaneous fluctuation. Based on the terrain feature information around the wind speed monitoring point, a spatial weight matrix is ​​generated to suppress the impact of local interference on the global pattern. Dynamic weight allocation step: based on the signal energy distribution of the steady-state component and the transient component and the historical matching confidence , calculate and distribute the matching weights of the steady-state component and the transient component in the subsequent pattern matching process in real time, wherein the historical matching confidence The number of successful matches within the preset time window based on the current mode Total matches The ratio of is determined, that is , and when the historical matching confidence is lower than a preset threshold, triggering the incremental learning update of the transient pattern library; Pattern library matching step: matching the candidate patterns that are most similar to the steady-state component and the transient component from the pre-built steady-state pattern library and the transient pattern library, respectively, and calculating the comprehensive matching degree in combination with the spatial weight matrix to determine the final wind speed monitoring identification result; Incremental learning and updating step: for transient signal features that are continuously mismatched, extract their feature information, and perform local iterative updating on the transient pattern library based on the extracted feature information. The local iterative updating includes comparing the transient signal features within the mismatch period with the existing transient pattern library for similarity. If the similarity is lower than a preset threshold, it is added as an independent pattern; otherwise, it is merged into the nearest neighbor pattern, and the steady-state pattern library remains unchanged. Among them, in the spatiotemporal feature decoupling step, the method for generating the spatial weight matrix includes: obtaining the height information and distribution density information of obstacles within a preset distance range around the wind speed monitoring point; based on the projection area of ​​the obstacles Angle from the current wind direction , dynamically adjust the attenuation coefficient of wind speed signals in each direction , the attenuation coefficient The angle There is a positive correlation.

2. The wind speed monitoring and identification method based on pattern matching according to claim 1 is characterized in that: In the dynamic weight allocation step, the preset time window is the past 30 seconds, and the preset threshold of the historical matching confidence is 0.

5.

3. The wind speed monitoring and identification method based on pattern matching according to claim 1 is characterized in that: In the spatiotemporal feature decoupling step, the real-time wind speed signal is decomposed by using lightweight Fourier transform to obtain the steady-state component and transient component.

4. The wind speed monitoring and identification method based on pattern matching according to claim 1 is characterized in that: In the pattern library matching step, the comprehensive matching degree is calculated on the embedded device by using a table lookup method instead of real-time matrix operation to reduce the calculation complexity.

5. The wind speed monitoring and identification method based on pattern matching according to claim 1 is characterized in that: After the real-time wind speed signal is decomposed into a steady-state component and a transient component in the spatiotemporal feature decoupling step, a morphological feature self-screening mechanism is also included, including: constructing a morphological template library, wherein the morphological template library is based on the transient signal waveform of historical turbulence events, extracting its local extreme value distribution, waveform symmetry and quantitative characteristics of energy attenuation slope to form no less than five turbulence templates; dynamically time-regularly matching the current transient signal with each template in the morphological template library to calculate the waveform morphological similarity; dynamically adjusting the morphological matching threshold according to the current ambient noise level, wherein the ambient noise level is obtained by inversely inferring the energy entropy of the steady-state component, and only retaining the candidate turbulence templates with a matching degree higher than the morphological matching threshold to enter the subsequent dynamic weight allocation process.

6. The wind speed monitoring and identification method based on pattern matching according to claim 5 is characterized in that: The method for constructing the morphological template library further includes a dynamic template self-organizing mechanism, which includes: automatically triggering the reconstruction of the morphological template library when the proportion of transient signal features that cannot be matched with the existing morphological template library in 10 consecutive incremental learning updates exceeds 30%; using a density-based online clustering algorithm to merge the newly added transient signal features and historical templates into superclasses according to morphological similarity, and deleting outdated templates that have not been matched in the past six months; and dynamically adjusting the weight ratio of various typical turbulence templates in the morphological template library based on the quarterly change trend of the steady-state component energy entropy.

7. The wind speed monitoring and identification method based on pattern matching according to claim 6 is characterized in that: In the dynamic template self-organizing mechanism, a sliding window density peak detection method is used for online clustering, retaining only the most representative morphological features in the current window and retaining no more than 20% of the morphological feature peak to control memory usage within a preset range.

8. The wind speed monitoring and identification method based on pattern matching according to claim 1 is characterized in that: The historical matching confidence is calculated by counting the ratio of the number of successful matches of the current pattern in the past 30 seconds to the total number of matches as the historical matching confidence. The method is applied to the fields of bridge structure health monitoring or meteorological environment monitoring.

9. A wind speed monitoring and identification system based on pattern matching, characterized in that: A system for implementing a pattern matching-based wind speed monitoring and identification method as described in any one of claims 1 to 8, the system comprising: a spatiotemporal feature decoupling module configured to decompose an acquired real-time wind speed signal into a steady-state component representing a long-term trend and a transient component representing an instantaneous fluctuation, and generate a spatial weight matrix for suppressing the influence of local interference on the global pattern based on terrain feature information around the wind speed monitoring point; a dynamic weight allocation module configured to calculate and allocate matching weights of the steady-state component and the transient component in a subsequent pattern matching process in real time based on the signal energy distribution of the steady-state component and the transient component and the historical matching confidence, wherein the historical matching confidence is determined based on the ratio of the number of successful matches of the current pattern within a preset time window to the total number of matches, and when the historical matching confidence is lower than a preset threshold, an incremental learning update of a transient pattern library is triggered; a pattern library matching module configured to match candidate patterns that are most similar to the steady-state component and the transient component from a pre-constructed steady-state pattern library and a pre-constructed transient pattern library, respectively, and calculate a comprehensive matching degree in combination with the spatial weight matrix to determine a final wind speed monitoring and identification result; The incremental learning update module is configured to extract characteristic information of transient signal features that are continuously mismatched, and perform local iterative updates on the transient pattern library based on the extracted characteristic information. The local iterative updates include performing similarity comparisons between the transient signal features within the mismatch period and the existing transient pattern library. If the similarity is lower than a preset threshold, the model is added as an independent model; otherwise, it is merged into the nearest neighbor model, and the steady-state pattern library remains unchanged.

Images