Intelligent correlation dimension identification method based on multi-window tracking and fuzzy C-means clustering method
Through multi-window tracking and fuzzy C-mean clustering method, the subjectivity and time-consuming problems of correlation dimension calculation in the existing technology are solved, and the rapid and accurate identification of correlation dimensions is achieved, and the calculation is improved. Efficiency and close to the truth.
Patent Information
- Application Number
- CN202411181043.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, the identification of scaleless intervals when calculating the correlation dimension of a time series depends on naked eye judgment, resulting in the calculation being too subjective and time-consuming, lacking objectivity, and difficult to accurately define under the influence of data noise.
The intelligent correlation dimension recognition method based on multi-window tracking and fuzzy C mean clustering method is adopted to reconstruct the phase space through the time delay embedding method, and the multi-window tracking technology is used to assist in determining the scale-free interval, and the fuzzy C mean clustering method is introduced Decompose the slope set, retain the slope fluctuation change part, take the average value as the correlation index, and combine the curve graphs of different hysteresis and embedded dimensions to determine the correlation dimension.
It realizes intelligent recognition of scaleless intervals, quickly and accurately obtains the associated dimension values, improves the calculation efficiency, and the obtained associated dimension values are closer to the true value, and overcomes the subjectivity and time-consuming problems of the original algorithm.
Smart Images

Figure CN119066458A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of correlation dimension recognition, and in particular relates to a correlation dimension intelligent recognition method based on multi-window tracking and fuzzy C-means clustering method. Background Art
[0002] Complexity is one of the most important measures in time series data analysis, and covers and is closely related to various other data characteristics in nonlinear system theory (such as stationarity, nonlinearity, fractality, periodicity, mutation, etc.). In the past few decades, many experts and scholars have devoted themselves to defining and quantifying complexity, and applied the theory of complexity to the study of natural and physical systems. Among them, the correlation dimension is an important indicator reflecting the complexity of the system. The larger the correlation dimension value, the more dominant processes the system needs, and the more complex its internal dynamic system is. Calculating the correlation dimension of a time series involves a variety of algorithms. The most widely used is the GP algorithm (proposed by Grassberger and Procaccia in 1983). The advantage of this algorithm is that it can handle high-dimensional data well and is very sensitive to small changes in the complexity of the dynamic structure of the data. The disadvantage is that it does not give the value range principle of the scalar r. Therefore, the identification of the scale-free interval (i.e., the linear part on the logC(r)~logr curve) can only be determined by the naked eye, which makes the calculation of the correlation dimension value too subjective. Theoretically, the scale-free interval is a straight line. However, due to the influence of data noise, there is no absolute linear part on the double logarithmic curve, so it takes a lot of time and effort to define the scale-free interval, which is a major obstacle to the application of this algorithm and a technical problem that needs to be solved urgently in this field. Based on this, the present invention proposes a correlation dimension intelligent identification method based on multi-window tracking and fuzzy C-means clustering method to realize intelligent identification of scale-free intervals, so as to quickly and accurately obtain correlation dimension values. Summary of the invention
[0003] The purpose of the present invention is to provide a correlation dimension intelligent recognition method based on multi-window tracking and fuzzy C-means clustering method to solve the above technical problems.
[0004] To achieve the above object, the present invention provides the following technical solutions:
[0005] The present invention discloses a correlation dimension intelligent recognition method based on multi-window tracking and fuzzy C-means clustering method, the method comprising the following steps:
[0006] Step 1: Reconstruction of phase space: Use the time delay embedding method to transform a time series x(t i )(i=1,2,3,...,n) is embedded into the m-dimensional phase space, and the m-dimensional vector time series Y(t i ):
[0007] Y(t i )={x(t i ),x(t i -τ),x(t i -2τ),...,x(t i -(m-1)τ)} (1)
[0008] Where i = 1, 2, 3, ..., n-(m-1)τ, τ is the lag time; m is the embedding dimension;
[0009] Step 2: Identification and calculation of scale-free intervals: The core of the correlation dimension GP algorithm is the correlation integral function, which is calculated as follows:
[0010]
[0011] Where H is the Heaviside step function, u=r-||Y(t a )-Y(t b )||; r is Y(t a ) or Y(t b ) is the radius of the sphere centered; ||Y(t a )-Y(t b )|| is the Euclidean distance between any two geometric points in the reconstructed phase space; N is the number of data points in the spatial sequence;
[0012] For a sufficiently long time series, the relationship between the correlation integral function C(r) and the radius r is:
[0013] C(r)≈αr v (3)
[0014] Among them, α is a constant; v is the correlation index;
[0015] The correlation index v is obtained by calculating the slope of the linear part of the logC(r)~logr curve, i.e. the scale-free interval:
[0016]
[0017] On the basis of using multi-window tracking technology to assist in determining the scale-free interval, a fuzzy C-means clustering method is introduced to divide the slope set into three parts: slope fluctuation change, slope continuous decline and slope saturation; the part of slope fluctuation change is retained, and the average value is the correlation index value under the corresponding lag τ and embedding dimension m; the slope set is obtained by using multi-window tracking technology to divide the logC(r)~logr curve into multiple segments, and using the least squares method to fit the data of each segment respectively;
[0018] Step 3: Determine the correlation dimension: Draw a graph of the correlation index v and the embedding dimension m at different lag times τ. When at least three curves reach saturation, take the average value of the saturation interval as the correlation dimension value of the time series.
[0019] Furthermore, the lag time τ in step 1 is in the range of 3 to 8, and the embedding dimension m is in the range of 3 to 20.
[0020] Furthermore, the sufficiently long time series in step 2 includes at least 300 data points.
[0021] Furthermore, the fuzzy C-means clustering method is introduced in step 2 to divide the slope set into three parts: slope fluctuation, slope continuous decline, and slope saturation; the part with slope fluctuation is retained, and the average value is the correlation index value under the corresponding lag τ and embedding dimension m, which specifically includes the following steps:
[0022] Step 21, determine the number of clusters C, initialize the membership matrix U = (u ij ) N×C ;
[0023] Among them, u ij ∈[0,1] is a random number, i=1,2,...,N, j=1,2,...,C, N is the number of elements in the slope set, and the constraints are satisfied
[0024] Step 22: In each iteration, the membership values in the membership matrix and the cluster centers of each group are continuously updated. The calculation formulas are:
[0025]
[0026] Among them, m is the fuzzy coefficient and is greater than 1; c j is the jth cluster center; x i is the i-th slope; c k is the kth cluster center; μ ij is the membership value of the jth cluster center corresponding to the ith slope; ||*|| is the Euclidean distance between two points;
[0027] The iteration termination condition is:
[0028] max ij {|μ ij (t+1) -μ ij (t) |}<ε (7)
[0029] Among them, ε is a constant, representing the error threshold;
[0030] Step 23: Select the part where the slope fluctuates, remove the first slope and the abnormal slope value and the slope before the abnormal slope value, and then take the average value of the remaining slopes to obtain the correlation index value under the corresponding lag time τ and embedding dimension m.
[0031] Furthermore, ε in step 22 is 10 -6 .
[0032] The beneficial effects of the present invention are as follows: the method for intelligent identification of correlation dimension based on multi-window tracking and fuzzy C-means clustering method described in the present invention overcomes the shortcomings of the correlation dimension GP algorithm that the identification of scale-free intervals needs to rely on naked eye identification, which consumes a lot of time and energy and lacks objectivity, realizes intelligent identification of scale-free intervals, can quickly and accurately obtain correlation dimension values, and the average computing efficiency is greatly improved compared with the original algorithm (R package nonlinearTseries). The calculated correlation dimension value is closer to the true value than the original algorithm, which is of great significance to the application of chaos theory.
[0033] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 is a flow chart of the method of the present invention;
[0035] Figure 2 This is the flow process diagram of Haba Gas Station in Daling River Basin from 1980 to 1999;
[0036] Figure 3 (Left) is the logC(r)~logr curve corresponding to the lag time of 6 and the embedding dimension of 10. The horizontal axis is the logarithm of the hypersphere radius, and the vertical axis is the logarithm of the correlation integral function;
[0037] Figure 3 (Middle) is the calculation diagram of the slope tracking of the segmented straight line, the horizontal axis is the number of segments, and the vertical axis is the slope;
[0038] Figure 3 (Right) is the result of fuzzy C-means clustering on the middle image, the horizontal axis is the number of segments, and the vertical axis is the slope;
[0039] Figure 4 It is the curve of correlation index v and embedding dimension m under different lag times τ. DETAILED DESCRIPTION
[0040] The present invention discloses a method for intelligently identifying correlation dimensions based on multi-window tracking and fuzzy C-means clustering method, such as Figure 1 As shown, the method comprises the following steps:
[0041] Step 1: Reconstruction of phase space: Use the time delay embedding method to transform a time series x(t i )(i=1,2,3,...,n) is embedded into the m-dimensional phase space, and the m-dimensional vector time series Y(t i ):
[0042] Y(t i )={x(t i ),x(t i -τ),x(t i -2τ),...,x(t i -(m-1)τ)} (1)
[0043] Where i=1,2,3,...,n-(m-1)τ, τ is the lag time, and m is the embedding dimension.
[0044] The lag τ and embedding dimension m are two important parameters in the correlation dimension GP algorithm. If their values are too large or too small, the final result of the correlation dimension will be affected. A large number of experiments have shown that the lag τ is usually between 3 and 8, and the embedding dimension m is usually between 3 and 20. Therefore, multiple sets of values are used for calculation.
[0045] Step 2: Identification and calculation of scale-free intervals: The core of the correlation dimension GP algorithm is the correlation integral function, which is calculated as follows:
[0046]
[0047] Where H is the Heaviside step function, u=r-||Y(t a )-Y(t b )||; r is Y(t a ) or Y(t b ) is the radius of the sphere centered; ||Y(t a )-Y(t b )|| is the Euclidean distance between any two geometric points in the reconstructed phase space; N is the number of data points in the spatial sequence.
[0048] For a sufficiently long time series (usually at least 300 data points), the relationship between the correlation integral function C(r) and the radius r is:
[0049] C(r)≈αr v (3)
[0050] Among them, α is a constant; v is the correlation index.
[0051] The correlation index v can be obtained by calculating the slope of the linear part (i.e., the scale-free interval) on the logC(r)~logr curve:
[0052]
[0053] Wherein, r and N have the same meanings as in formula (2).
[0054] Due to the emergence of data noise, the linear part (i.e., the scale-free interval) on the logC(r)-logr curve is difficult to define. The traditional method is to determine it by visual estimation. The present invention introduces the fuzzy C-means clustering method on the basis of using multi-window tracking technology to assist in determining the scale-free interval, and divides the slope set into three parts: slope fluctuation change, slope continuous decline, and slope saturation. The slope set is obtained by using multi-window tracking technology to perform multi-segment segmentation on the logC(r)-logr curve, and using the least squares method to fit the data of each segment respectively; retain the part of the slope fluctuation change, and take the average value as the correlation index value under the corresponding lag τ and embedding dimension m, which specifically includes the following three steps:
[0055] Step 21, determine the number of clusters C, initialize the membership matrix U = (u ij ) N×C ;
[0056] Among them, u ij ∈[0,1] is a random number, i=1,2,...,N, j=1,2,...,C, N is the number of elements in the slope set, and the constraints are satisfied
[0057] Step 22: In each iteration, the membership values in the membership matrix and the cluster centers of each group are continuously updated. The calculation formulas are as follows:
[0058]
[0059] Among them, m is the fuzzy coefficient and is greater than 1; c j is the jth cluster center; x i is the i-th slope; c k is the kth cluster center; μ ij is the membership value of the jth cluster center corresponding to the ith slope; ||*|| is the Euclidean distance between two points.
[0060] The iteration termination condition is:
[0061] max ij {|μ ij (t+1) -μ ij (t) |}<ε (7)
[0062] Here, ε is a very small constant representing the error threshold, usually 10 -6.
[0063] Step 23: Select the part where the slope fluctuates, remove the first slope and the abnormal slope value and the slope before the abnormal slope value, and then average the remaining slopes to obtain the correlation index value under the corresponding lag time τ and embedding dimension m.
[0064] Step 3, determination of correlation dimension: In order to ensure the reliability and robustness of the estimated correlation dimension value, at least three correlation index v~embedding dimension m curves with different lag times τ are required to reach saturation. Here, the lag time τ is an integer between 3 and 8, and the embedding dimension m is an integer between 3 and 20. The correlation index v~embedding dimension m curves with different lag times τ are plotted. When at least three curves reach saturation, the average value of the saturation interval is taken as the correlation dimension (CD) value of the time series.
[0065] Embodiment 1
[0066] This embodiment takes the flow sequence of Haba Gas Station in Daling River Basin from 1980 to 1999 as an example to illustrate the application of the above method in practice.
[0067] This embodiment discloses a correlation dimension intelligent identification method based on multi-window tracking and fuzzy C-means clustering method, comprising the following steps:
[0068] Step 1: Reconstruction of phase space: The time delay embedding method is used to reconstruct the phase space of the flow sequence of Haba gas station in Daling River Basin from 1980 to 1999. Figure 2 As shown, we get the m-dimensional vector flow sequence Y(t i ).
[0069] Step 2: Identification and calculation of scale-free intervals: Taking the lag τ as 6 and the embedding dimension m as 10 as an example, calculate the correlation integral function according to formula (2) and draw the logC(r)~logr curve, as shown in Figure 3 (left) As shown; using multi-window tracking technology, the logC(r)~logr curve is divided into 50 small segments for tracking, and the least squares method is used to fit the data of each small segment to obtain the slope set and draw a slope change curve, as shown in Figure 3 (middle); using the fuzzy C-means clustering method, the slope set is divided into three parts: slope fluctuation, slope continuous decline and slope saturation, as shown in Figure 3 As shown in (right), the part of the slope fluctuation is retained, the first slope and the abnormal slope value and the slope before the abnormal slope value are eliminated, and the average value of the remaining slope is calculated to be 2.541, which is the correlation index corresponding to the lag τ of 6 and the embedding dimension m of 10.
[0070] Step 3: Determination of correlation dimension: Here, the lag time τ is an integer between 3 and 8, and the embedding dimension m is an integer between 3 and 20. Plot the correlation index v ~ embedding dimension m curve at different lag times τ, as shown in Figure 4 As shown. When at least three curves reach saturation, the correlation dimension (CD) value fluctuates between 2.555 and 2.568, and the final correlation dimension value is 2.56, which is closer to the true value (i.e., 2.75) than the value calculated by the original algorithm (R package nonlinearTseries) (i.e., 2.49). The true value is calculated by manual visual estimation combined with multi-window tracking technology, which takes a very long time, and it takes at least 3 hours to calculate one value.
[0071] Finally, it should be noted that the above is only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred arrangement scheme, a person skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.
Claims
1. An intelligent identification method of correlation dimension based on multi-window tracking and fuzzy C-means clustering method, It is characterized in that The method comprises the following steps: Step 1: Reconstruction of phase space: Use the time delay embedding method to transform a time series x(t i )(i=1,2,3,...,n) is embedded into the m-dimensional phase space, and the m-dimensional vector time series Y(t i ): Y(t i )={x(t i ),x(t i -τ),x(t i -2τ),...,x(t i -(m-1)τ)} (1) Where i = 1, 2, 3, ..., n-(m-1)τ, τ is the lag time; m is the embedding dimension; Step 2: Identification and calculation of scale-free intervals: The core of the correlation dimension GP algorithm is the correlation integral function, which is calculated as follows: Where H is the Heaviside step function, u=r-||Y(t a )-Y(t b )||; r is Y(t a ) or Y(t b ) is the radius of the sphere centered; ||Y(t a )-Y(t b )|| is the Euclidean distance between any two geometric points in the reconstructed phase space; N is the number of data points in the spatial sequence; For a sufficiently long time series, the relationship between the correlation integral function C(r) and the radius r is: C(r)≈αr v (3) Where α is a constant; v is the correlation index; The correlation index v is obtained by calculating the slope of the linear part of the logC(r)~logr curve, i.e. the scale-free interval: On the basis of using multi-window tracking technology to assist in determining the scale-free interval, a fuzzy C-means clustering method is introduced to divide the slope set into three parts: slope fluctuation change, slope continuous decline and slope saturation; the part of slope fluctuation change is retained, and the average value is the correlation index value under the corresponding lag τ and embedding dimension m; the slope set is obtained by using multi-window tracking technology to divide the logC(r)~logr curve into multiple segments, and using the least squares method to fit the data of each segment respectively; Step 3: Determine the correlation dimension: Draw a graph of the correlation index v and the embedding dimension m at different lag times τ. When at least three curves reach saturation, take the average value of the saturation interval as the correlation dimension value of the time series.
2. According to claim 1, a correlation dimension intelligent recognition method based on multi-window tracking and fuzzy C-means clustering method, It is characterized in that The lag time τ in step 1 ranges from 3 to 8, and the embedding dimension m ranges from 3 to 20.
3. According to claim 1, a correlation dimension intelligent recognition method based on multi-window tracking and fuzzy C-means clustering method, It is characterized in that The sufficiently long time series in step 2 includes at least 300 data points.
4. According to claim 1, a correlation dimension intelligent recognition method based on multi-window tracking and fuzzy C-means clustering method, It is characterized in that In step 2, the fuzzy C-means clustering method is introduced to divide the slope set into three parts: slope fluctuation, slope continuous decline, and slope saturation; The part where the slope fluctuates is retained, and the average value is taken as the correlation index value under the corresponding lag τ and embedding dimension m, which specifically includes the following steps: Step 21, determine the number of clusters C, initialize the membership matrix U = (u ij ) N×C ; Among them, u ij ∈[0,1] is a random number, i=1,2,...,N, j=1,2,...,C, N is the number of elements in the slope set, and the constraints are satisfied Step 22: In each iteration, the membership values in the membership matrix and the cluster centers of each group are continuously updated. The calculation formulas are: Among them, m is the fuzzy coefficient and is greater than 1; c j is the jth cluster center; x i is the i-th slope; c k is the kth cluster center; μ ij is the membership value of the jth cluster center corresponding to the i-th slope; ||*|| is the Euclidean distance between two points; The iteration termination condition is: max ij {|m ij (t+1) -m ij (t) |}<e (7) Among them, ε is a constant, representing the error threshold; Step 23: Select the part where the slope fluctuates, remove the first slope and the abnormal slope value and the slope before the abnormal slope value, and then take the average value of the remaining slopes to obtain the correlation index value under the corresponding lag time τ and embedding dimension m.
5. According to claim 4, a correlation dimension intelligent recognition method based on multi-window tracking and fuzzy C-means clustering method, It is characterized in that ε in step 22 is 10 -6 .
Citation Information
Patent Citations
Hydrological model parameter optimization method based on correlation dimension
CN110647717A
Error threshold determination method for detecting electric energy quality monitoring device by using fractal dimensions
CN113884969A
