A rail transit intelligent monitoring system
By performing feature decomposition and adaptive Gaussian filtering processing on rail transit data, the monitoring inaccurate problem caused by fixed Gaussian filtering scales in the prior art is solved, and more accurate rail transit data monitoring is achieved.
Patent Information
- Application Number
- CN202411081624.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-08-08
AI Technical Summary
The existing Gaussian filtering method adopts a fixed Gaussian scale, which leads to inaccurate monitoring results of rail transit data and ineffective denoising.
By performing feature decomposition of the settlement data of rail transit nodes, feature components are obtained, and the Gaussian filter core size is adaptively adjusted according to the degree of discreteness and noise distribution characteristic value of the power spectrum of each feature component, and filtering is performed to remove noise.
It improves the noise removal effect, makes the rail transit data monitoring results more accurate and reliable, and can flexibly process the complex structure of signals, enhancing the accuracy of monitoring.
Smart Images

Figure CN118606635B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to an intelligent monitoring system for rail transit. Background Art
[0002] As a key component of the urban transportation network, the rail transit system is essential for safe, comfortable and efficient operation. However, over time, the problem of surface subsidence has become a serious challenge facing the rail transit system. Surface subsidence may cause damage to infrastructure such as tracks, tunnels, and bridges, which in turn affects the safety of train operation and the reliability of the rail system. In order to promptly detect and deal with surface subsidence problems, the existing rail transit monitoring system is often used to collect track subsidence data in real time and monitor it. The main monitoring steps include data preprocessing, data cleaning, data extraction, data conversion, data mining, data analysis, etc.
[0003] Among them, the data preprocessing in the monitoring step is mainly to filter the noise data contained in the collected original settlement data, so as to make the monitoring results of the monitoring system more accurate. The existing Gaussian filter is often used to filter the signal components of the original settlement data after EMD empirical mode decomposition, but the existing Gaussian filter method often uses a fixed Gaussian scale, which makes the filtering result less accurate, and thus the rail transit data monitoring result is less accurate. Summary of the invention
[0004] In order to solve the technical problem that the existing Gaussian filtering method often adopts a fixed Gaussian scale, which makes the filtering result less accurate, and thus makes the rail transit data monitoring result less accurate, the purpose of the present invention is to provide a rail transit intelligent monitoring system, and the technical solution adopted is as follows:
[0005] Obtain the settlement data of rail transit nodes at different times within a set time period, perform feature decomposition on the settlement data to obtain feature components; filter the feature components according to the discrete degree of data in the power spectrum corresponding to each feature component to obtain the components to be denoised;
[0006] According to the data distribution of each component to be denoised at the same time, the data fluctuation of each component to be denoised, and the frequency fluctuation of each component to be denoised within a local time range, the noise distribution characteristic value at each moment is obtained;
[0007] According to the noise distribution characteristic value at each moment and the noise distribution of the data in the power spectrum corresponding to each component to be denoised at each moment, the comprehensive noise level at each moment is obtained;
[0008] The Gaussian filter kernel size at each moment is obtained according to the comprehensive noise level, and each component to be denoised at each moment is filtered respectively using the Gaussian filter kernel size, and the filtered components are used to monitor rail transit.
[0009] Preferably, the method of screening the characteristic components according to the discrete degree of data in the power spectrum corresponding to each characteristic component to obtain the components to be denoised specifically includes:
[0010] For any characteristic component, a power spectrum of the characteristic component is obtained, and the data discreteness of the characteristic component is obtained according to the difference distribution between the power value at each moment and the power value at all moments in the power spectrum of the characteristic component, and the difference distribution between the frequency value at each moment and the frequency value at all moments;
[0011] The characteristic components are screened according to the degree of data dispersion to obtain the components to be denoised.
[0012] Preferably, the data discreteness of the characteristic component is obtained according to the difference distribution between the power value at each moment and the power value at all moments in the power spectrum of the characteristic component, and the difference distribution between the frequency value at each moment and the frequency value at all moments, specifically including:
[0013] Record any characteristic component as the selected characteristic component, calculate the mean value of the power value of the selected characteristic component at all times and record it as the power characteristic value of the selected characteristic component, calculate the mean value of the frequency value of the selected characteristic component at all times and record it as the frequency characteristic value of the selected characteristic component;
[0014] The first difference at each moment is obtained by calculating the absolute value of the difference between the power value at each moment and the power characteristic value; the second difference at each moment is obtained by calculating the absolute value of the difference between the frequency value at each moment and the frequency characteristic value; the product of the first difference and the second difference is used as the data difference degree at each moment; the normalized value of the mean of the data difference degrees at all moments is calculated to obtain the data discreteness degree of the selected characteristic component.
[0015] Preferably, the step of screening the characteristic components according to the degree of data dispersion to obtain the components to be denoised specifically includes:
[0016] The characteristic components corresponding to the data discreteness greater than the preset degree threshold are taken as the components to be denoised.
[0017] Preferably, obtaining the noise distribution characteristic value at each moment according to the data distribution of each component to be denoised at the same moment, the data fluctuation of each component to be denoised, and the frequency fluctuation of each component to be denoised within a local time range specifically includes:
[0018] Record any moment as the target moment, and obtain the first characteristic coefficient according to the data amplitude of each component to be denoised at the target moment and the fluctuation of all data amplitudes in each component to be denoised;
[0019] The difference between the frequency values of each component to be denoised within the local time range of the target time and the fluctuation of all the frequency values in each component to be denoised are used to obtain the second characteristic coefficient;
[0020] A noise distribution characteristic value at a target moment is obtained according to the first characteristic coefficient and the second characteristic coefficient.
[0021] Preferably, the method for obtaining the noise distribution characteristic value is specifically:
[0022]
[0023] in, represents the noise distribution characteristic value at the target time, t represents the t-th time, represents the total number of components to be denoised, represents the data amplitude of the i-th component to be denoised at the t-th time, Represents the variance of all data amplitudes of the i-th component to be denoised; It represents the rth frequency value excluding the target time in the window with length L and centered at the tth time in the i-th component to be denoised. represents the frequency value at the tth moment in the i-th component to be denoised, represents the variance of all frequency values in the i-th component to be denoised, Represents the product of the variances of the data amplitudes of all components to be denoised, represents the product of the variances of the frequency values of all components to be denoised, and Norm( ) represents the linear normalization function.
[0024] Preferably, the method of obtaining the comprehensive noise level at each moment according to the noise distribution characteristic value at each moment and the noise distribution of the data in the power spectrum corresponding to each component to be denoised at each moment specifically includes:
[0025] Record any moment as the selected moment, perform normalization processing on the mean of all frequency values in the power spectrum of all components to be denoised at the selected moment, and obtain a first coefficient; perform negative correlation normalization processing on the mean of all power values in the power spectrum of all components to be denoised at the selected moment, and obtain a second coefficient; obtain the initial noise level at the selected moment according to the first coefficient and the second coefficient; the first coefficient and the second coefficient are both positively correlated with the initial noise level;
[0026] The product of the initial noise level at the selected moment and the noise distribution characteristic value is normalized to obtain the comprehensive noise level at the selected moment.
[0027] Preferably, obtaining the Gaussian filter kernel size at each moment according to the comprehensive noise level specifically includes:
[0028]
[0029] in, represents the Gaussian filter kernel size at the tth moment, represents the preset initial filter scale, represents the comprehensive noise level at the tth moment, Indicates the round-up symbol.
[0030] Preferably, the step of performing feature decomposition on the settlement data to obtain feature components specifically includes:
[0031] The IMF components are obtained by performing EMD decomposition on the settlement data, and all the settlement data within the set time period constitute the original data sequence; the correlation coefficient between each IMF component and the original data sequence is calculated, and the IMF component corresponding to the correlation coefficient greater than the preset correlation threshold is taken as the characteristic component.
[0032] Preferably, the monitoring of rail transit using the filtered components specifically includes:
[0033] The components to be denoised after filtering, other characteristic components and residual items obtained by EMD decomposition are recombined, and the recombined data are used to monitor rail transit.
[0034] The embodiments of the present invention have at least the following beneficial effects:
[0035] The present invention first obtains the settlement data of the rail transit node, and obtains the characteristic components through characteristic decomposition, which enables denoising to be performed on different frequency scales, thereby more flexibly processing the complex structure of the signal and achieving better denoising effect. In addition, by analyzing the discrete degree of the data in the power spectrum corresponding to each characteristic component, the characteristic components are preliminarily screened to obtain the components to be denoised, so that the subsequent denoising operation can be more targeted and the denoising effect is better. Then, by analyzing the data distribution of each component to be denoised at the same time and the data fluctuation of each component to be denoised, as well as the frequency fluctuation of each component to be denoised within the local time range, the noise performance at each moment is quantified to obtain the noise distribution characteristic value, that is, the noise performance degree of the data at each moment is characterized. Further, the noise distribution is further quantified by combining the noise distribution characteristic value and the noise distribution of the data in the power spectrum corresponding to each component to be denoised at each moment, and the comprehensive noise degree is used to characterize the comprehensive noise performance of the data at each moment. Finally, the Gaussian filter kernel size at each moment can be adaptively obtained through the comprehensive noise performance of the data, so that the filtering effect of the component to be denoised is better, and the rail transit data monitoring results are more accurate. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0037] Figure 1 It is a flowchart of the steps of a method executed by a rail transit intelligent monitoring system provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0038] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation method, structure, features and effects of a rail transit intelligent monitoring system proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0039] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0040] A specific scheme of a rail transit intelligent monitoring system provided by the present invention is described in detail below with reference to the accompanying drawings.
[0041] This embodiment proposes a rail transit intelligent monitoring system to achieve the following Figure 1 The specific steps are as follows:
[0042] Step 1: Obtain the settlement data of rail transit nodes at different times within a set time period, perform feature decomposition on the settlement data to obtain feature components; filter the feature components according to the discrete degree of data in the power spectrum corresponding to each feature component to obtain the components to be denoised.
[0043] In the rail transit system, the importance of different sections is different. For important nodes in rail transit, structural settlement sensors are usually installed on the tracks of important nodes to measure the settlement data at the important nodes of the train track. Among them, important nodes can be transfer nodes and intersection nodes of train tracks. Settlement data refers to the measurement data of the vertical displacement of the track in the rail transit system relative to its original position. Settlement data is used to monitor and evaluate the stability and safety of rail infrastructure.
[0044] In this embodiment, the settlement data measured at any node position is used as an example to obtain the settlement data of the node at different times within a set time period, set the length of the set time period to 1 day, and set the time interval between two adjacent moments within the set time period to 1 minute. The implementer can set it according to the specific implementation scenario. It should be noted that the settlement data of the track should be collected and monitored for no less than three months.
[0045] The collected settlement data can be decomposed into multiple IMF components through EMD decomposition. Each IMF component corresponds to a local frequency in the original data, which allows denoising to be performed on different frequency scales, thereby more flexibly processing the complex structure of the signal and achieving better denoising effects. That is, the settlement data is subjected to feature decomposition to obtain feature components, and the correlation between the multiple IMF components obtained by decomposition and the original settlement data is compared to identify multiple IMF components containing the main information.
[0046] Specifically, the settlement data is decomposed by EMD to obtain IMF components, and all the settlement data within a set time period constitute the original data sequence; the correlation coefficient between each IMF component and the original data sequence is calculated, and the IMF component corresponding to the correlation coefficient greater than the preset correlation threshold is taken as the characteristic component.
[0047] In this embodiment, the correlation coefficient is obtained by calculating the covariance between each IMF component and the original data sequence. The larger the value of the covariance between the IMF component and the original data sequence, the stronger the correlation between the two sets of data, and the larger the value of the corresponding correlation coefficient. Among them, the value of the correlation threshold is 0.4. The larger the value of the correlation threshold, the stricter the standard for screening the IMF component, and the smaller the value of the correlation threshold, the looser the standard for screening the IMF component.
[0048] When the correlation coefficient corresponding to the IMF component is greater than the correlation threshold, it means that the corresponding IMF component contains more main information and has a greater correlation with the original data sequence. When the correlation coefficient corresponding to the IMF component is less than or equal to the correlation threshold, it means that the corresponding IMF component contains less main information and has a smaller correlation with the original data sequence. Based on this, the acquired characteristic component represents the IMF component with a greater correlation with the original data sequence. At the same time, the residual term data is also obtained by performing EMD decomposition on the sedimentation data.
[0049] After filtering out the characteristic components that contain more main information in the original data, the power spectrum of each characteristic component can be used to determine whether each characteristic component is mixed with noise data, and targeted filtering can be performed. That is, according to the discrete degree of data in the power spectrum corresponding to each characteristic component, the characteristic components are filtered to obtain the components to be denoised.
[0050] First, the discrete degree of the data in the power spectrum corresponding to each characteristic component is analyzed, that is, for any characteristic component, the power spectrum of the characteristic component is obtained, and the data discrete degree of the characteristic component is obtained according to the difference distribution between the power value at each moment and the power value at all moments in the power spectrum of the characteristic component, and the difference distribution between the frequency value at each moment and the frequency value at all moments. Among them, the method of obtaining the power spectrum is a well-known technology and will not be introduced here.
[0051] Specifically, any characteristic component is recorded as the selected characteristic component, the mean of the power values of the selected characteristic component at all times is calculated and recorded as the power characteristic value of the selected characteristic component, and the mean of the frequency values of the selected characteristic component at all times is calculated and recorded as the frequency characteristic value of the selected characteristic component; the absolute value of the difference between the power value at each moment and the power characteristic value is calculated to obtain the first difference at each moment; the absolute value of the difference between the frequency value at each moment and the frequency characteristic value is calculated to obtain the second difference at each moment; the product of the first difference and the second difference is taken as the data difference degree at each moment; the normalized value of the mean of the data difference degree at all moments is calculated to obtain the data discreteness degree of the selected characteristic component.
[0052] In this embodiment, the nth characteristic component is taken as an example for explanation, and the nth characteristic component is taken as the selected characteristic component. Then, the calculation formula of the data discreteness of the selected characteristic component can be expressed as:
[0053]
[0054] in, Indicates the data discreteness of the nth characteristic component, Indicates the total number of included moments, Represents the frequency value at the tth moment in the power spectrum of the nth characteristic component, Represents the mean of the frequency values at all times in the power spectrum of the nth characteristic component, that is, the frequency characteristic value of the nth characteristic component; represents the power value at the tth moment in the power spectrum of the nth characteristic component, It represents the mean value of the power values at all times in the power spectrum of the nth eigencomponent, that is, the power eigenvalue of the nth eigencomponent, and Norm( ) represents the linear normalization function.
[0055] It is the first difference at the t-th moment, reflecting the difference between the power value of the n-th characteristic component at the t-th moment and the overall value, and reflecting the outlier of the power data corresponding to the characteristic component. is the second difference at the tth moment, reflecting the difference between the frequency value of the nth characteristic component at the tth moment and the whole, and reflecting the outlier of the frequency data corresponding to the characteristic component. The larger the value of the first difference and the second difference, the greater the degree of outlier of the data at each moment in the characteristic component, the greater the corresponding data discreteness, and the greater the value of the data discreteness.
[0056] It should be noted that in this embodiment, when calculating the degree of data discreteness, both frequency and power are calculated using corresponding numerical values. Since the track settlement data has a small discreteness under normal circumstances, that is, there is no high-frequency fluctuation of its own, and the frequency data of the original settlement data is relatively stable. However, the noise data is relatively high-frequency and distributed in various frequency ranges. Therefore, in each characteristic component, the greater the deviation between the local frequency data and the overall frequency data, the greater the deviation between the local power data and the overall power data, indicating that the noise content in the characteristic component is higher.
[0057] Based on this, the characteristic components are screened according to the degree of data discreteness to obtain the components to be denoised. Specifically, the characteristic components corresponding to the data discreteness greater than the preset degree threshold are used as the components to be denoised. In this embodiment, the value of the degree threshold is set to 0.8, and the implementer can set it according to the specific implementation scenario. Since the degree of data discreteness is a normalized value, the value of the degree threshold is (0,1). When the value of the degree threshold is closer to 1, it means that the standard for noise screening of the characteristic components is stricter. When the value of the degree threshold is closer to 0, it means that the standard for noise screening of the characteristic components is looser.
[0058] When the data discreteness of the feature component is greater than the degree threshold, it means that the greater the data outlier in the feature component, the greater the corresponding discreteness, and thus the greater the noise contained in the feature component. When the data discreteness of the feature component is less than or equal to the degree threshold, it means that the data outlier of the feature component value is smaller, the corresponding discreteness is smaller, and thus the noise contained in the feature component is smaller. Based on this, the component to be denoised represents the part of the feature component with a larger noise degree.
[0059] Step 2: According to the data distribution of each component to be denoised at the same time, the data fluctuation of each component to be denoised, and the frequency fluctuation of each component to be denoised within a local time range, the noise distribution characteristic value at each moment is obtained.
[0060] In the track settlement data, the faster the change of the data amplitude in the IMF component is relative to the overall change, the more abnormal the IMF data change is, and the more likely the IMF component data is noise data. However, when analyzing the abnormal changes in the IMF component, it is also necessary to consider the impact changes in the data. When the data has an impact change, it means that the abnormal change in the data may be caused by a sudden change in the data, and the degree of data noise at this time is also greater.
[0061] In this embodiment, the peak factor of the data is expressed by the sum of the data amplitudes in the component to be denoised. The larger the factor, the greater the degree of change of the data in the component to be denoised, and the more likely it is the noise performance of the sedimentation data. At the same time, the impact change of the data will also be reflected in the frequency, that is, within the local time range of each moment, the greater the distribution difference between the frequency of each other data and the frequency of the central moment, the more likely the data at the central moment is the noise performance of the sedimentation data.
[0062] Based on this, the noise distribution at each moment is quantified by analyzing the data distribution of each component to be denoised at the same moment, the data fluctuation of each component to be denoised, and the frequency fluctuation of each component to be denoised within a local time range.
[0063] Specifically, any moment is recorded as the target moment, and the first characteristic coefficient is obtained according to the data amplitude of each component to be denoised and the fluctuation of all data amplitudes in each component to be denoised at the target moment; the second characteristic coefficient is obtained according to the difference between the frequency values of each component to be denoised and the fluctuation of all frequency values in each component to be denoised within the local time range of the target moment; the noise distribution characteristic value at the target moment is obtained according to the first characteristic coefficient and the second characteristic coefficient.
[0064] In this embodiment, the tth moment is taken as the target moment, and the calculation formula of the noise distribution characteristic value at the target moment can be expressed as:
[0065]
[0066] in, represents the noise distribution characteristic value at the target time, t represents the t-th time, represents the total number of components to be denoised, represents the data amplitude of the i-th component to be denoised at the t-th time, Represents the variance of all data amplitudes of the i-th component to be denoised; It represents the rth frequency value excluding the target time in the window with length L and centered at the tth time in the i-th component to be denoised. represents the frequency value at the tth moment in the i-th component to be denoised, represents the variance of all frequency values in the i-th component to be denoised, Represents the product of the variances of the data amplitudes of all components to be denoised, represents the product of the variances of the frequency values of all components to be denoised, and Norm( ) represents the linear normalization function.
[0067] It represents the cumulative sum of the data amplitudes of the signal components containing more noise at the same time, reflecting the sudden changes in the settlement data at the target time. It reflects the fluctuation of the data amplitude of each component to be denoised. It reflects the prominence of the data amplitude in each signal component containing more noise at the target time. The larger the value is, the greater the degree of change of the sedimentation data at the target time, and the more likely it is an impact change of the sedimentation data. The corresponding noise situation at the target time is greater, that is, the larger the value of the noise distribution characteristic value is.
[0068] In this embodiment, the data within the local time range of the target moment is obtained in each component to be denoised at the target moment, and the difference between the data within the local time range is analyzed. For example, in the i-th component to be denoised, with the t-th moment as the center, L moments are obtained as the local time range of the t-th moment, where the value of L is 7. It indicates the difference between other frequency values and the frequency value at the target time within the local time range of the target time on the i-th component to be denoised, reflecting the change of local frequency in the component to be denoised. The larger the value, The larger the value of is, the greater the difference between the frequency value and the surrounding frequency value in each denoised component is. It reflects the fluctuation of the frequency value of each component to be denoised. The larger the value of is, the more abnormal the frequency data at the target time is, and the more likely it is an impact change in the settlement data. That is, the greater the degree of noise in the corresponding data, the larger the value of the noise distribution characteristic value.
[0069] The noise distribution characteristic value at each moment represents the noise level of the data at each moment, and reflects the mutation of the data amplitude and frequency value at each moment. The larger the value, the greater the noise level at the corresponding moment.
[0070] Step three, according to the noise distribution characteristic value at each moment and the noise distribution of the data in the power spectrum corresponding to each component to be denoised at each moment, the comprehensive noise level at each moment is obtained.
[0071] Since the track settlement data fluctuates less under normal circumstances, the smaller the fluctuation of the data frequency in the characteristic component, the smaller the fluctuation of the data power. Then, the initial noise distribution at each moment can be quantified by combining the distribution of the data frequency and data power values in each component to be denoised at each moment. Then, the quantized initial noise distribution can be combined with the noise distribution characteristic value at the corresponding moment, which can more accurately characterize the degree of data noise at each moment.
[0072] Specifically, any moment is recorded as the selected moment, and the mean of all frequency values in the power spectrum of all components to be denoised at the selected moment is normalized to obtain a first coefficient; the mean of all power values in the power spectrum of all components to be denoised at the selected moment is negatively normalized to obtain a second coefficient; the initial noise level at the selected moment is obtained according to the first coefficient and the second coefficient; the first coefficient and the second coefficient are both positively correlated with the initial noise level.
[0073] In this embodiment, the sth moment is taken as the selected moment, and the calculation formula of the initial noise level at the selected moment can be expressed as:
[0074]
[0075] in, represents the initial noise level at the selected moment, represents the total number of components to be denoised, Represents the frequency value at the sth moment in the power spectrum of the i-th component to be denoised; represents the power value at the sth moment in the power spectrum of the i-th component to be denoised, Norm( ) represents the linear normalization function, and exp( ) represents the exponential function with the natural constant e as the base.
[0076] is the first coefficient. The larger the data frequency at the selected time, the faster the data fluctuation is, and the less likely it is a normal data fluctuation. The stronger the noise performance at the selected time, the larger the corresponding initial noise level value is.
[0077] is the second coefficient. The greater the data power performance at the selected moment, the wider the data distribution range in the denoising component at the selected moment, and the less likely it is noise data. In other words, the weaker the noise performance at the selected moment, the smaller the corresponding initial noise level value.
[0078] The initial noise level at each moment reflects the initial noise distribution in all components to be denoised at each moment. Then the product of the initial noise level at the selected moment and the noise distribution eigenvalue is normalized to obtain the comprehensive noise level at the selected moment. The comprehensive noise level combines the initial noise distribution at each moment and the mutation of the data at each moment, and more comprehensively reflects the comprehensive noise distribution in all components to be denoised at each moment.
[0079] Step 4: Obtain the Gaussian filter kernel size at each moment according to the comprehensive noise level, use the Gaussian filter kernel size to filter each component to be denoised at each moment, and use the filtered components to monitor rail transit.
[0080] The larger the value of the comprehensive noise level at each moment, the stronger the data noise performance at the corresponding moment, and the larger the scale should be used for noise reduction at that moment. The smaller the value of the comprehensive noise level at each moment, the weaker the data noise performance at the corresponding moment, and the smaller the scale should be used for noise reduction at that moment.
[0081] Based on this, the Gaussian filter kernel size at each moment is obtained according to the comprehensive noise level, that is, the initial Gaussian filter kernel size needs to be set in advance, and the initial Gaussian filter kernel size can be adaptively adjusted using the comprehensive noise level at each moment, which can be expressed as:
[0082]
[0083] in, represents the Gaussian filter kernel size at the tth moment, represents the preset initial filter scale, represents the comprehensive noise level at the tth moment, Indicates the round-up symbol. In this embodiment, The value of is 5.
[0084] The larger the value of the comprehensive noise level at the tth moment, the stronger the noise performance is, and the larger the value of the corresponding adjusted Gaussian filter kernel size is. The smaller the value of the comprehensive noise level at the tth moment, the weaker the noise performance is, and the smaller the value of the corresponding adjusted Gaussian filter kernel size is.
[0085] At this point, the Gaussian filter kernel size adaptively obtained at each moment can be used to denoise each component to be denoised at each moment, and the denoised components to be denoised can be combined with other feature components that have not been denoised and the residual terms obtained by EMD decomposition to perform signal recombination, and the recombined data can be used to monitor rail transit.
[0086] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.
Claims
1. A rail transit intelligent monitoring system, comprising a memory and a processor, characterized in that: The processor executes the computer program stored in the memory to implement the following steps: Obtain the settlement data of rail transit nodes at different times within a set time period, perform feature decomposition on the settlement data to obtain feature components; filter the feature components according to the discrete degree of data in the power spectrum corresponding to each feature component to obtain the components to be denoised; According to the data distribution of each component to be denoised at the same time, the data fluctuation of each component to be denoised, and the frequency fluctuation of each component to be denoised within a local time range, the noise distribution characteristic value at each moment is obtained; According to the noise distribution characteristic value at each moment and the noise distribution of the data in the power spectrum corresponding to each component to be denoised at each moment, the comprehensive noise level at each moment is obtained; According to the comprehensive noise level, the Gaussian filter kernel size is obtained at each moment, each component to be denoised at each moment is filtered by using the Gaussian filter kernel size, and the filtered components are used to monitor rail transit; The noise distribution characteristic value at each moment is obtained according to the data distribution of each component to be denoised at the same moment, the data fluctuation of each component to be denoised, and the frequency fluctuation of each component to be denoised within a local time range, specifically including: Record any moment as the target moment, and obtain the first characteristic coefficient according to the data amplitude of each component to be denoised at the target moment and the fluctuation of all data amplitudes in each component to be denoised; The difference between the frequency values of each component to be denoised within the local time range of the target time and the fluctuation of all the frequency values in each component to be denoised are used to obtain the second characteristic coefficient; Obtaining a noise distribution characteristic value at a target time according to the first characteristic coefficient and the second characteristic coefficient; The method for obtaining the noise distribution characteristic value is specifically as follows: in, represents the noise distribution characteristic value at the target time, t represents the t-th time, represents the total number of components to be denoised, represents the data amplitude of the i-th component to be denoised at the t-th time, Represents the variance of all data amplitudes of the i-th component to be denoised; It represents the rth frequency value excluding the target time in the window with length L and centered at the tth time in the i-th component to be denoised. represents the frequency value at the tth moment in the i-th component to be denoised, represents the variance of all frequency values in the i-th component to be denoised, Represents the product of the variances of the data amplitudes of all components to be denoised, represents the product of the variances of the frequency values of all components to be denoised, and Norm( ) represents the linear normalization function.
2. The rail transit intelligent monitoring system according to claim 1, characterized in that: The method of screening the characteristic components according to the discrete degree of data in the power spectrum corresponding to each characteristic component to obtain the components to be denoised specifically includes: For any characteristic component, a power spectrum of the characteristic component is obtained, and the data discreteness of the characteristic component is obtained according to the difference distribution between the power value at each moment and the power value at all moments in the power spectrum of the characteristic component, and the difference distribution between the frequency value at each moment and the frequency value at all moments; The characteristic components are screened according to the degree of data dispersion to obtain the components to be denoised.
3. A rail transit intelligent monitoring system according to claim 2, characterized in that: The data discreteness of the characteristic component is obtained according to the difference distribution between the power value at each moment and the power value at all moments in the power spectrum of the characteristic component, and the difference distribution between the frequency value at each moment and the frequency value at all moments, specifically including: Record any characteristic component as the selected characteristic component, calculate the mean value of the power value of the selected characteristic component at all times and record it as the power characteristic value of the selected characteristic component, calculate the mean value of the frequency value of the selected characteristic component at all times and record it as the frequency characteristic value of the selected characteristic component; The first difference at each moment is obtained by calculating the absolute value of the difference between the power value at each moment and the power characteristic value; the second difference at each moment is obtained by calculating the absolute value of the difference between the frequency value at each moment and the frequency characteristic value; the product of the first difference and the second difference is used as the data difference degree at each moment; the normalized value of the mean of the data difference degrees at all moments is calculated to obtain the data discreteness degree of the selected characteristic component.
4. The rail transit intelligent monitoring system according to claim 2, characterized in that: The step of screening the characteristic components according to the degree of data dispersion to obtain the components to be denoised specifically includes: The characteristic components corresponding to the data discreteness greater than the preset degree threshold are taken as the components to be denoised.
5. The rail transit intelligent monitoring system according to claim 2, characterized in that: The method of obtaining the comprehensive noise level at each moment according to the noise distribution characteristic value at each moment and the noise distribution of the data in the power spectrum corresponding to each component to be denoised at each moment specifically includes: Record any moment as the selected moment, perform normalization processing on the mean of all frequency values in the power spectrum of all components to be denoised at the selected moment, and obtain a first coefficient; perform negative correlation normalization processing on the mean of all power values in the power spectrum of all components to be denoised at the selected moment, and obtain a second coefficient; obtain the initial noise level at the selected moment according to the first coefficient and the second coefficient; the first coefficient and the second coefficient are both positively correlated with the initial noise level; The product of the initial noise level at the selected moment and the noise distribution characteristic value is normalized to obtain the comprehensive noise level at the selected moment.
6. The rail transit intelligent monitoring system according to claim 1, characterized in that: The obtaining of the Gaussian filter kernel size at each moment according to the comprehensive noise level specifically includes: in, represents the Gaussian filter kernel size at the tth moment, represents the preset initial filter scale, represents the comprehensive noise level at the tth moment, Indicates the round-up symbol.
7. The rail transit intelligent monitoring system according to claim 1, characterized in that: The characteristic decomposition of the settlement data to obtain characteristic components specifically includes: The IMF components are obtained by performing EMD decomposition on the settlement data, and all the settlement data within the set time period constitute the original data sequence; the correlation coefficient between each IMF component and the original data sequence is calculated, and the IMF component corresponding to the correlation coefficient greater than the preset correlation threshold is taken as the characteristic component.
8. The rail transit intelligent monitoring system according to claim 7, characterized in that: The monitoring of rail transit by using the filtered components specifically includes: The components to be denoised after filtering, other characteristic components and residual items obtained by EMD decomposition are recombined, and the recombined data are used to monitor rail transit.
