Daily suspended matter concentration profile change identification method, device and computer program

Abstract

The application discloses a daily suspended matter concentration structure change identification method, equipment and computer program, the method constructs the time series of daily suspended matter concentration observation data, carries out multiscale decomposition to the time series, obtains a plurality of intrinsic mode function (IMF) components and a residual term; the average oscillation period of each IMF component is calculated respectively, the components exceeding the preset period threshold are automatically identified, and the components are removed from the original time series to construct an initial background sequence; further, the components with the period falling into the preset window are identified from the remaining components, and a reconstructed background sequence is constructed in combination with the initial background sequence, trend item and seasonal item decomposition and structure change breakpoint identification are carried out on the reconstructed background sequence, and finally, the structure change stage division and mutation time output of the daily suspended matter concentration time series are realized. The method of the application can improve the accuracy and robustness of the structure change detection of the daily suspended matter concentration time series.

Description

Technical Field

[0001] This invention belongs to the field of time series analysis technology of water environment elements, specifically involving a method, device and computer program for identifying daily suspended solids concentration structure changes. Background Technology

[0002] Time series analysis of suspended particulate matter concentration typically employs statistical methods such as linear regression analysis, moving average analysis, Mann-Kendall trend test, and abrupt change test. These methods primarily identify long-term trends or abrupt changes in the time series through overall statistical characteristics, and are effective in handling stationary time series. However, daily suspended particulate matter concentration time series often exhibit significant non-stationarity and non-linearity, while also being superimposed with high-frequency fluctuations and extreme disturbances. In such cases, traditional statistical methods struggle to effectively distinguish between long-term trends and short-term fluctuations, and are easily affected by outliers or random noise, leading to unstable abrupt change identification results or misjudgments.

[0003] To enhance the analysis capabilities of non-stationary signals, Empirical Mode Decomposition (EMD) and its improved algorithms have been increasingly applied to hydrological time series research, including methods such as Ensemble Empirical Mode Decomposition (EEMD) and Fully Adaptive Noise Ensemble Empirical Mode Decomposition (CEEMDAN). These methods can decompose complex time series into several intrinsic mode function components, thereby revealing the variation characteristics at different time scales and offering advantages in signal denoising and multi-scale analysis. However, existing multi-scale decomposition methods are typically only used for signal component analysis or trend interpretation, lacking the ability to directly identify abrupt changes in the time series structure.

[0004] On the other hand, structural abrupt change detection methods can identify the locations of structural changes in time series. For example, methods such as Breaks For Additive Seasonal and Trend (BFAST) are applied in identifying structural changes in time series. These methods can identify breakpoints in time series through trend and seasonal decomposition. However, when these methods are directly applied to raw daily suspended matter concentration time series, they are easily affected by high-frequency fluctuations and anomalous disturbances, resulting in false breakpoints or unstable identification results.

[0005] Therefore, existing technologies still lack a technical method that can identify stable structural mutations based on multi-scale decomposition, in order to improve the accuracy and robustness of detecting daily suspended matter concentration time series structural changes. Summary of the Invention

[0006] The purpose of this invention is to provide a method, device, and computer program for identifying daily suspended particulate matter concentration structural changes, in order to solve the problem that existing suspended particulate matter concentration time series are easily affected by high-frequency fluctuations and abnormal disturbances, resulting in unstable breakpoint results and unclear stage characteristics when identifying structural changes, thereby improving the accuracy, robustness, and interpretability of daily suspended particulate matter concentration structural change identification.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for identifying daily changes in suspended matter concentration structure, the method comprising:

[0009] A time series of daily suspended particulate matter concentration observation data is constructed, and the time series is decomposed into multiple scales to obtain several intrinsic mode function (IMF) components and residual terms.

[0010] Calculate the average oscillation period of the IMF components and extract the components whose average oscillation period is less than a preset period threshold; add the extracted components one by one according to their time positions to form a first component sequence; subtract the first component sequence from the time series of the daily suspended matter concentration observation data one by one according to their time positions to construct an initial background sequence;

[0011] For the remaining IMF components that have not been extracted, identify the components whose average oscillation period falls within a preset annual cycle window, and subtract the components falling within the preset annual cycle window from the initial background sequence point by point according to the time position to obtain the reconstructed background sequence.

[0012] The reconstructed background sequence is decomposed into trend and seasonal terms, structural change breakpoints in the trend terms are identified, and the structural change stages of the suspended matter concentration time series are divided based on the structural change breakpoints.

[0013] In some embodiments of the present invention, the time series of the daily suspended matter concentration observation data is as follows:

[0014] The year, month, and date fields in the suspended matter concentration observation data are combined to form a standard date sequence, and the suspended matter concentration variable data are arranged in chronological order of date to form a time series of suspended matter concentration observation data within the study period;

[0015] Invalid date records are deleted from the time series, and missing years or data gaps are marked; during multi-scale decomposition and structural change breakpoint identification, continuous time periods are extracted for processing.

[0016] In some embodiments of the present invention, the time series of the daily suspended matter concentration observation data is decomposed into multiple scales using the CEEMDAN method.

[0017] In some embodiments of the present invention, the average oscillation period is estimated based on the zero-crossing characteristics of the IMF components; IMF components with an average oscillation period less than a preset period threshold are identified as high-frequency components.

[0018] In some embodiments of the present invention, the preset period threshold is set to 30 days; the preset annual period window is set to 300-420 days.

[0019] In some embodiments of the present invention, if multiple IMF components fall within the preset annual cycle window, the single IMF component closest to the annual target cycle is identified; or,

[0020] Multiple IMF components falling within the preset annual cycle window are superimposed as the identified component.

[0021] In some embodiments of the present invention, the reconstructed background sequence is used as input, and the BFAST model is used to decompose the trend term and seasonal term, and the structural change breakpoints in the trend term are identified.

[0022] In some embodiments of the present invention, the time series is divided into multiple change stages according to the structural change breakpoints, and the mean, fluctuation characteristics and change trends of each stage are statistically analyzed to characterize the structural differences in the changes of suspended matter concentration in different stages.

[0023] The present invention further provides a computer device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0024] The present invention further provides a computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0025] This invention addresses the problem that various disturbances in daily suspended particulate matter concentration time series can easily mask structural changes and lead to unstable breakpoint identification. It constructs a processing flow for identifying structural abrupt changes in suspended particulate matter concentration time series. While preserving seasonal variations and long-term trend information, it reduces the interference of disturbances on the structural abrupt change identification results, thereby achieving accurate and stable identification of daily suspended particulate matter concentration structural changes.

[0026] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below may be considered part of the inventive subject matter of this disclosure, provided that such concepts do not contradict each other. Furthermore, all combinations of the claimed subject matter are considered part of the inventive subject matter of this disclosure.

[0027] The foregoing and other aspects, embodiments, and features of the teachings of the present invention will be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the invention, such as features and / or beneficial effects of exemplary embodiments, will become apparent from the following description or may be learned through practice of specific embodiments according to the teachings of the present invention. Attached Figure Description

[0028] The accompanying drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component shown in the various figures may be denoted by the same reference numeral. For clarity, not every component is labeled in each figure. Embodiments of various aspects of the invention will now be described by way of example and with reference to the accompanying drawings, wherein:

[0029] Figure 1 This is a technical roadmap for a method to identify the daily suspended matter concentration time series structural changes based on high-frequency stripping, provided for embodiments of the present invention.

[0030] Figure 2 This is a graph showing the daily suspended matter concentration data of Datong Station in an embodiment of the present invention.

[0031] Figure 3 This is a schematic diagram illustrating the statistical analysis of missing years in daily suspended matter observation data in an embodiment of the present invention.

[0032] Figure 4 This is a partial CEEMDAN multiscale decomposition result of the daily suspended matter concentration time series at Datong Station in an embodiment of the present invention.

[0033] Figure 5 This is a diagram showing the identification results of the average oscillation period of each IMF component in an embodiment of the present invention.

[0034] Figure 6 This is a comparison chart of the original daily suspended matter concentration time series and the background series in an embodiment of the present invention, where the gray line represents the original series and the black line represents the initial background series.

[0035] Figure 7 This is a comparison diagram of the BFAST input sequence (reconstructed background sequence) and the initial background sequence in an embodiment of the present invention, where the gray line represents the initial background sequence and the black line represents the reconstructed background sequence.

[0036] Figure 8 This is a comparison diagram of the initial background sequence, annual cycle component, and reconstructed background sequence in an embodiment of the present invention.

[0037] Figure 9 This is a diagram showing the BFAST structural mutation identification results in an embodiment of the present invention, using the reconstructed background sequence as input.

[0038] In the aforementioned Figures 1-9, the coordinates, symbols, or other representations expressed in English are all well-known in the field and will not be elaborated upon in this example. Detailed Implementation

[0039] To better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.

[0040] Various aspects of the invention are described in this disclosure with reference to the accompanying drawings, which illustrate numerous illustrative embodiments. The embodiments of this disclosure are not necessarily intended to encompass all aspects of the invention. It should be understood that the various concepts and embodiments described above, as well as those described in more detail below, can be implemented in any of many ways, as the concepts and embodiments disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.

[0041] Example 1

[0042] This embodiment uses daily suspended matter observation data from a hydrological station from 2003 to 2024 as an example to illustrate the method for identifying daily suspended matter concentration structure changes described in this invention. The daily suspended matter concentration data used is employed to analyze the synergy and rationality of the suspended matter change process. The specific process of the method is as follows: Figure 1 As shown, the specific steps include the following:

[0043] Step 1: Obtain raw daily suspended matter observation data

[0044] Obtain daily suspended particulate matter (SPM) observation data from a hydrological station from 2003 to 2024. The data includes at least the year, month, date, and SPM concentration. Combine the year, month, and date fields to form a standard date sequence, and arrange the original data in chronological order to form a daily SPM observation sequence for the study period.

[0045] In this embodiment, the original daily suspended matter concentration observation data are as follows: Figure 2 As shown.

[0046] Step 2: Preprocess the raw data

[0047] The daily suspended matter observation data are preprocessed, and the preprocessing includes:

[0048] (1) Check whether the year, month, and date fields can form a valid date;

[0049] (2) Delete data rows with invalid dates or incorrect records;

[0050] (3) Identify missing years within the research period;

[0051] (4) Count the number of valid observation records in each year.

[0052] In this embodiment, missing years are not manually filled in, nor is the overall year range cropped. Instead, the original research period is retained, and broken lines are displayed at the corresponding positions of the missing years in subsequent graphical representations. The statistical results of missing years are as follows: Figure 3 As shown. When there are missing years or data gaps within the study period, subsequent CEEMDAN decomposition and BFAST structural mutation identification can be performed separately for consecutive time periods.

[0053] Step 3: Construct daily time series of suspended solids concentrations

[0054] After data preprocessing, daily suspended particulate matter concentration time series were obtained. These daily suspended particulate matter concentration time series were used as the target analysis object in this embodiment for subsequent multi-scale decomposition and structural change identification.

[0055] Step 4: Perform CEEMDAN multiscale decomposition on the daily suspended matter concentration time series.

[0056] The daily suspended particulate matter concentration time series constructed in step 3 is input into the CEEMDAN model for multi-scale decomposition, decomposing the original daily suspended particulate matter concentration time series into several intrinsic mode functions IMF1, IMF2, ..., IMF 𝑘 Each IMF component represents the oscillation characteristics of the original sequence at different time scales. The earlier modes typically correspond to short-period fluctuations, while the later modes correspond to mid-to-low frequency changes and long-term background information.

[0057] In this embodiment, the CEEMDAN model ensemble degree is set to 100, the noise intensity coefficient is set to 0.2, and multiple intrinsic mode functions (IMF0~IMF0) are obtained after decomposing the daily suspended matter concentration time series. 11 ); Figure 4 The diagram shows a portion (IMF6~IMF9 and the residuals).

[0058] Step 5: Automatically identify high-frequency components

[0059] Calculate the average oscillation period for each IMF component obtained in step 4, such as... Figure 5 The system automatically identifies high-frequency disturbance components based on a preset period threshold. In this embodiment, IMF components with an average oscillation period of less than 30 days are identified as high-frequency components.

[0060] The identified high-frequency IMF components are summed hourly according to their time positions to form a high-frequency component sequence with the same length as the original daily suspended matter concentration time series. Let the set of IMF indices identified as high-frequency components be denoted as . Then the high-frequency component sequence can be represented as:

[0061]

[0062] in, This represents a high-frequency component sequence, where each component in the sequence... for: The high-frequency perturbation value is formed by adding the high-frequency IMF components at each time step. Using this method, a one-dimensional high-frequency component sequence characterizing the short-period perturbation features of the original sequence can be constructed, providing a foundation for the subsequent construction of the initial background sequence.

[0063] In this embodiment, the high-frequency component extraction results are shown in Table 1.

[0064] Table 1. Results of High-Frequency Component Extraction (Partial Excerpt)

[0065] 2003 / 3 / 9 0.143 0.176 -0.0331 2004 / 4 / 1 0.155 0.125 0.0291 2005 / 2 / 22 0.307 0.225 0.0814 2006 / 5 / 20 0.186 0.144 0.0411 2007 / 8 / 5 0.378 0.423 -0.0455 2008 / 8 / 27 0.363 0.325 0.0374 2009 / 8 / 27 0.218 0.250 -0.032 2010 / 5 / 25 0.229 0.196 0.032 2011 / 9 / 5 0.099 0.071 0.0276 2012 / 7 / 18 0.371 0.299 0.0715

[0066] Step 6: Remove high-frequency components and construct the initial background sequence

[0067] The high-frequency component sequence is obtained in step 5. Subsequently, the high-frequency component sequence was extracted from the original daily suspended matter concentration time series. The initial background sequence is obtained by peeling away the material point by point according to its time position. Its expression is:

[0068]

[0069] Due to the high-frequency component sequence Since the high-frequency IMF components are summed point-by-point according to time, the above stripping process is mathematically equivalent to reconstructing the remaining mid- and low-frequency IMF components and residual terms that were not identified as high-frequency components. Initial background sequence It can also be expressed as:

[0070]

[0071] in, This is a set of high-frequency IMF component indices. This refers to the residual terms obtained from the CEEMDAN decomposition.

[0072] The initial background sequence constructed in this way mainly retains the characteristics of low- and medium-frequency fluctuations, seasonal changes, and long-term background changes, while weakening the impact of short-term high-frequency random disturbances, local abnormal fluctuations, and noise components on the trend identification results.

[0073] The initial background sequence obtained in step 6 is used to characterize the mid-to-low frequency variation background after stripping away high-frequency perturbations, and the results are as follows: Figure 6 As shown. Figure 6 In the image, the gray line represents the original daily suspended matter concentration sequence, and the black line represents the initial background sequence.

[0074] It should be noted that this is based on high-frequency component sequences. Obtaining the background sequence by subtracting it from the original sequence time-by-time is mathematically equivalent to reconstructing the background sequence directly using mid-to-low frequency IMF components that were not identified as high-frequency components and residual terms. The former is used to characterize the specific process of high-frequency stripping, while the latter is used to explain the compositional origin of the background sequence.

[0075] Step 7: Identify the annual cycle components and reconstruct the background sequence

[0076] Based on the high-frequency component identification results from step 5, or the decomposition results corresponding to the background sequence obtained in step 6, annual cycle components are identified in the IMF components that are not identified as high-frequency components, according to whether their average oscillation period falls within a preset cycle window. In this embodiment, the preset cycle window is 300 to 420 days. If multiple IMF components fall within the annual cycle window, the IMF component closest to the 365-day target cycle is selected as a single annual cycle component, or multiple related IMF components are superimposed as an annual cycle component. The identified annual cycle component is used to characterize the annual-scale regularity of changes in the daily suspended matter concentration time series, and can be used to analyze the periodic fluctuations, peak and trough positions, and cycle strength characteristics of suspended matter concentration within the year. By distinguishing the annual cycle component from the reconstructed background sequence, the separation of annual cycle change information from long-term structural change information can be achieved, thereby providing support for the construction of the reconstructed background sequence and subsequent identification of structural abrupt changes.

[0077] After identifying the annual cycle component, the low-frequency IMF components (excluding the annual cycle component) and the residual term are merged to form the reconstructed background sequence. In other words, the initial background sequence is composed of the annual cycle component and the reconstructed background sequence, which satisfies:

[0078]

[0079] in, This represents the initial background sequence obtained in step 6. This indicates the identified annual cycle component. This indicates the reconstruction of the background sequence.

[0080] In this embodiment, the annual cycle component It can be represented as:

[0081]

[0082] in, This represents the set of annual cycle IMF component serial numbers obtained from the identification.

[0083] Reconstructing the background sequence It can be represented as:

[0084]

[0085] in, This represents the set of low-frequency IMF component indices excluding the annual cycle component. This represents the residual terms obtained from the CEEMDAN decomposition.

[0086] The obtained reconstructed background sequence is used as the input sequence for subsequent BFAST structural change identification.

[0087] Step 8: Reconstruct the background sequence BFAST trend breakpoint identification was performed, and the identification results were obtained.

[0088] The reconstructed background sequence from step 7 is input into the BFAST model. Trend-seasonal decomposition is performed on the reconstructed background sequence, and structural change breakpoints in the trend term are identified. The reconstructed background sequence, used as the BFAST input sequence, is decomposed by the BFAST model into a trend term, a seasonal term, and a stochastic residual term, as follows:

[0089]

[0090] in, For trend items, For seasonal items, This is a random residual term.

[0091] In this embodiment, the minimum segment length ratio of the BFAST model is set to 0.15, the seasonal term form is set to harmonic, and the maximum number of iterations is set to 10. By performing trend fitting and piecewise regression on the reconstructed background sequence, trend breakpoints in the daily suspended matter concentration time series are identified, resulting in a trend breakpoint identification result map based on the method of this invention, as shown in the figure. Figure 9 As shown.

[0092] Figure 7The relationship between the initial background sequence and the BFAST input sequence (reconstructed background sequence) is demonstrated. In this invention, the reconstructed background sequence, which mainly characterizes the low-frequency changes and long-term background changes in the daily suspended matter concentration time series, is used as the input parameter of the BFAST model. This can reduce the interference of short-term high-frequency fluctuations on the structural change identification results while preserving seasonal changes and long-term trend information.

[0093] Figure 8 The relationship between the initial background sequence, the annual cycle component, and the reconstructed background sequence is illustrated. The initial background sequence can be decomposed into two parts: the annual cycle component and the reconstructed background sequence. The annual cycle component characterizes the annual-scale regularity of changes in daily suspended matter concentration time series, while the reconstructed background sequence characterizes the long-term changes after removing high-frequency disturbances and annual cycle fluctuations. By explicitly identifying the annual cycle component, intra-annual periodic fluctuations can be distinguished from long-term structural changes, providing a more targeted input sequence for subsequent identification of structural change breakpoints.

[0094] Figure 9 The results demonstrate the identification of structural change breakpoints obtained by using the reconstructed background sequence as the input sequence for BFAST. These results show that the present invention can identify long-term structural changes in daily suspended matter concentration time series and output the locations of structural change breakpoints and corresponding stage division results.

Claims

1. A method for identifying daily changes in suspended matter concentration structure, characterized in that, The method includes: A time series of daily suspended particulate matter concentration observation data is constructed, and the time series is decomposed into multiple scales to obtain several intrinsic mode function (IMF) components and residual terms. Calculate the average oscillation period of the IMF components and extract the components whose average oscillation period is less than a preset period threshold; add the extracted components one by one according to their time positions to form a first component sequence; subtract the first component sequence from the time series of the daily suspended matter concentration observation data one by one according to their time positions to construct an initial background sequence; For the remaining IMF components that have not been extracted, identify the components whose average oscillation period falls within a preset annual cycle window, and subtract the components falling within the preset annual cycle window from the initial background sequence point by point according to the time position to obtain the reconstructed background sequence. The reconstructed background sequence is decomposed into trend and seasonal terms, structural change breakpoints in the trend terms are identified, and the structural change stages of the suspended matter concentration time series are divided based on the structural change breakpoints.

2. The method according to claim 1, characterized in that, The time series of the daily suspended matter concentration observation data is as follows: The year, month, and date fields in the suspended matter concentration observation data are combined to form a standard date sequence, and the suspended matter concentration data are arranged in chronological order to form a time series of suspended matter concentration observation data within the study period. Invalid date records are deleted from the time series, and missing years or data gaps are marked; during multi-scale decomposition and structural change breakpoint identification, continuous time periods are extracted for processing.

3. The method according to claim 1, characterized in that, The time series of the daily suspended matter concentration observation data was decomposed into multiple scales using the CEEMDAN method.

4. The method according to claim 1, characterized in that, The average oscillation period is estimated based on the zero-crossing characteristic of the IMF components.

5. The method according to claim 1, characterized in that, The preset period threshold is set to 30 days; the preset annual period window is set to 300-420 days.

6. The method according to claim 1, characterized in that, If multiple IMF components fall within the preset annual cycle window, then identify the single IMF component closest to the annual target cycle; or, Multiple IMF components falling within the preset annual cycle window are superimposed as the identified component.

7. The method according to claim 1, characterized in that, Using the reconstructed background sequence as input, the BFAST model is used to decompose the trend term and seasonal term, and the structural change breakpoints in the trend term are identified.

8. The method according to claim 1, characterized in that, The time series is divided into multiple change stages based on the structural change breakpoints, and the mean, fluctuation characteristics and change trends of each stage are statistically analyzed to characterize the structural differences in suspended matter concentration changes in different stages.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 8.

Images