1. A wind turbine generator tower attitude prediction method is characterized by comprising the following steps:

acquiring multi-element time sequence data of the wind turbine tower attitude;

performing phase space reconstruction on the multivariate time series data by using an improved C-C method;

constructing a plurality of corresponding initial prediction models for the multivariate time sequence data after the phase space reconstruction; the initial prediction model is a support vector regression model;

optimizing each initial prediction model by adopting a sequential minimum method;

training each optimized prediction model based on the multivariate time series data after the phase space reconstruction;

and predicting the attitude of the tower drum of the wind turbine through the trained prediction models.

2. The wind turbine tower attitude prediction method of claim 1, wherein the multivariate time series data comprises angular velocity of the wind turbine tower along the x-axis, the y-axis, and the z-axis, and an inclination angle of the wind turbine tower along the x-axis, the y-axis, and the z-axis.

3. The wind turbine tower attitude prediction method of claim 1, wherein before performing the phase-space reconstruction of the multivariate time series data using an improved C-C method, further comprising:

and carrying out normalization processing on the multivariate time series data.

4. The method for predicting the attitude of a tower of a wind turbine generator as claimed in claim 1, wherein the kernel function of the support vector regression model is a radial basis function.

5. The wind turbine tower attitude prediction method according to claim 1, wherein the training of each optimized prediction model based on the multivariate time series data after the phase space reconstruction specifically comprises:

inputting the multivariate time sequence data after the phase space reconstruction into each optimized prediction model to obtain output data;

and optimizing the penalty factor and the nuclear parameter in each optimized prediction model based on the output data to complete the training process.

6. The wind turbine tower attitude prediction method according to claim 1, wherein the performing phase-space reconstruction on the multivariate time series data by using an improved C-C method specifically comprises:

determining delay time of each variable in the multivariate time sequence data;

calculating an embedding dimension;

performing a phase space reconstruction of the multivariate time series data based on the delay time and the embedding dimension.

7. A wind turbine tower attitude prediction system, comprising:

the multivariate time sequence data acquisition module is used for acquiring multivariate time sequence data of the wind turbine tower attitude;

the phase space reconstruction module is used for performing phase space reconstruction on the multivariate time sequence data by utilizing an improved C-C method;

the model construction module is used for constructing a plurality of corresponding initial prediction models for the multivariate time series data after the phase space reconstruction; the initial prediction model is a support vector regression model;

the optimization module is used for optimizing each initial prediction model by adopting a sequential minimum method;

the training module is used for training each optimized prediction model based on the multi-element time sequence data after the phase space reconstruction;

and the prediction module is used for predicting the attitude of the tower drum of the wind turbine through the trained prediction models.

8. The wind turbine tower attitude prediction system of claim 7, wherein the training module specifically comprises:

the input unit is used for inputting the multivariate time sequence data after the phase space reconstruction into each optimized prediction model to obtain output data;

and the optimization unit is used for optimizing the penalty factors and the nuclear parameters in each optimized prediction model based on the output data to complete the training process.

9. The wind turbine tower attitude prediction system of claim 7, wherein the phase space reconstruction module specifically comprises:

a delay time determining unit for determining a delay time of each variable in the multivariate time series data;

an embedding dimension calculation unit for calculating an embedding dimension;

and the phase space reconstruction unit is used for performing phase space reconstruction on the multivariate time sequence data based on the delay time and the embedding dimension.

Background

With the rapid development of science and technology, modern engineering systems become more and more complex, and the role of the modern engineering systems is more and more important. Complex systems generally work in a harsh environment, and because the system improves performance and increases uncertain factors influencing normal operation of the system, the potential possibility of system failure or malfunction is greatly increased. Often, a minor failure of a component will cause the performance of the whole system to gradually decrease, and even a failure of a critical component will cause the system to fail to operate properly or even to fail. Many catastrophic failures are caused by minor failures and result in significant property loss, casualties and social impact. In order to ensure that no fatal fault occurs in the operation process of a complex system, the requirements on the reliability and the safety of each subsystem and subsystem are extremely high. The reliability of the whole system is improved by improving the reliability of components and assemblies, and the requirement of a complex system on the reliability is difficult to meet, so the fault diagnosis and fault-tolerant control technology becomes an important technology for improving the reliability and the safety of the complex system.

The wind turbine installation and rotation platform online monitoring system is a closed-loop system with complex characteristics, and generally has the problems of data shortage, difficult establishment of a prediction model, serious uncertainty and the like, so that the fault prediction of the system has many difficulties, such as how to establish the prediction model, how to establish a prediction mechanism, how to ensure the prediction accuracy and real-time performance and the like, which need to be subjected to long-term deep analysis and discussion. In summary, the research on the failure prediction theory and method is a necessary way, and has a general meaning, the results of theoretical research and practical application of the method can bring endless benefits to complex systems with extremely high requirements on reliability and safety, and the research results can also be popularized and applied to various national economic fields, such as carrier rockets, space stations, satellites, manned spacecrafts, other complex industrial systems and the like, and have wide application prospects.

Disclosure of Invention

The invention aims to provide a wind turbine generator tower attitude prediction method and system, which are used for predicting the attitude of a wind turbine tower accurately and in real time so as to realize the prediction of faults.

In order to achieve the purpose, the invention provides the following scheme:

a wind turbine generator tower attitude prediction method comprises the following steps:

acquiring multi-element time sequence data of the wind turbine tower attitude;

performing phase space reconstruction on the multivariate time series data by using an improved C-C method;

constructing a plurality of corresponding initial prediction models for the multivariate time sequence data after the phase space reconstruction; the initial prediction model is a support vector regression model;

optimizing each initial prediction model by adopting a sequential minimum method;

training each optimized prediction model based on the multivariate time series data after the phase space reconstruction;

and predicting the attitude of the tower drum of the wind turbine through the trained prediction models.

Further, the multivariate time series data includes angular velocity of the wind turbine tower along the x-axis, the y-axis, and the z-axis, and an inclination angle of the wind turbine tower along the x-axis, the y-axis, and the z-axis.

Further, before performing a phase space reconstruction on the multivariate time series data by using an improved C-C method, the method further comprises:

and carrying out normalization processing on the multivariate time series data.

Further, the kernel function of the support vector regression model is a radial basis function.

Further, the training of each optimized prediction model based on the phase space reconstructed multivariate time series data specifically includes:

inputting the multivariate time sequence data after the phase space reconstruction into each optimized prediction model to obtain output data;

and optimizing the penalty factor and the nuclear parameter in each optimized prediction model based on the output data to complete the training process.

Further, the performing phase space reconstruction on the multivariate time series data by using the improved C-C method specifically includes:

determining delay time of each variable in the multivariate time sequence data;

calculating an embedding dimension;

performing a phase space reconstruction of the multivariate time series data based on the delay time and the embedding dimension.

The invention also provides a wind turbine generator tower attitude prediction system, which comprises:

the multivariate time sequence data acquisition module is used for acquiring multivariate time sequence data of the wind turbine tower attitude;

the phase space reconstruction module is used for performing phase space reconstruction on the multivariate time sequence data by utilizing an improved C-C method;

the model construction module is used for constructing a plurality of corresponding initial prediction models for the multivariate time series data after the phase space reconstruction; the initial prediction model is a support vector regression model;

the optimization module is used for optimizing each initial prediction model by adopting a sequential minimum method;

the training module is used for training each optimized prediction model based on the multi-element time sequence data after the phase space reconstruction;

and the prediction module is used for predicting the attitude of the tower drum of the wind turbine through the trained prediction models.

Further, the training module specifically includes:

the input unit is used for inputting the multivariate time sequence data after the phase space reconstruction into each optimized prediction model to obtain output data;

and the optimization unit is used for optimizing the penalty factors and the nuclear parameters in each optimized prediction model based on the output data to complete the training process.

Further, the phase space reconstruction module specifically includes:

a delay time determining unit for determining a delay time of each variable in the multivariate time series data;

an embedding dimension calculation unit for calculating an embedding dimension;

and the phase space reconstruction unit is used for performing phase space reconstruction on the multivariate time sequence data based on the delay time and the embedding dimension.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects:

the time sequence is processed by utilizing a phase space reconstruction method, and a weighted SVR prediction model is established, namely, a sequential minimum method is adopted to optimize each support vector regression model, and meanwhile, the shutdown criterion and the model parameters are optimized; therefore, the attitude of the wind turbine tower can be predicted more accurately, and the fault prediction is realized.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.

FIG. 1 is a flow chart of a wind turbine tower attitude prediction method according to an embodiment of the invention;

FIG. 2 is a schematic diagram illustrating the principle of optimizing each of the initial prediction models by a sequential minimization method according to an embodiment of the present invention; a drawing;

FIG. 3 is an optimal parameter fitness curve for a first set of variables;

FIG. 4 is an optimal parameter fitness curve for a second set of variables;

FIG. 5 is an optimal parameter fitness curve for a third set of variables;

FIG. 6 is an optimal parameter fitness curve for a fourth set of variables;

FIG. 7 is an optimal parameter fitness curve for a fifth set of variables;

FIG. 8 is an optimal parameter fitness curve for a sixth set of variables;

FIG. 9 is a one-step prediction of 22s to 32s angular velocity p for the onset of a fault;

FIG. 10 is a one-step prediction of angular velocity q from 22s to 32s of fault onset;

FIG. 11 shows the angular velocities r from 22s to 32s at which a fault starts₀Predicting the result in one step;

FIG. 12 is a one-step prediction of the fault onset 22s to 32s angle φ;

FIG. 13 is a graph of the results of a one-step prediction of the 22s to 32s angle θ for the onset of a fault;

FIG. 14 is a one-step prediction of the 22s to 32s angle ψ for the onset of a fault;

FIG. 15 shows the results of a one-step prediction of the angular velocity p to be predicted;

FIG. 16 is a one-step prediction result of the angular velocity q to be predicted;

FIG. 17 shows the angular velocity r to be predicted₀Predicting the result in one step;

FIG. 18 shows the results of one-step prediction of the angle φ to be predicted

FIG. 19 shows the one-step prediction result of the angle θ to be predicted

Fig. 20 shows the result of one-step prediction of the angle ψ to be predicted.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The invention aims to provide a wind turbine generator tower attitude prediction method and system, which are used for predicting the attitude of a wind turbine tower accurately and in real time so as to realize the prediction of faults.

To achieve the above object, first, the present invention uses Lyapunov exponent to represent the motion behavior of chaotic attractors. Wherein, the time sequence refers to an observed value of a certain variable arranged at equal time intervals according to the time sequence; the Lyapunov exponent method can measure the average convergence or average divergence of the system between adjacent tracks in a phase space, is sensitive to specific initial conditions and has strong immunity to noise; chaotic attractors in phase space can describe chaotic systems, attractors are finite regions that the systems tend to in phase space, and chaotic attractors are attractors with self-similar structures formed by tracks of different levels. Furthermore, the invention adopts a small data method to solve the maximum Lyapunov index of each state variable time sequence of the fan tower attitude online monitoring system, thereby judging the wonderness of the system. Wherein the ith Lyapunov index is defined as:

i＝1,2,…,n

wherein n represents an n-dimensional phase space, p_i(n) represents the spindle length.

According to the formula, the angular rates p, q and r of the fan tower barrel along the x axis, the y axis and the z axis can be respectively calculated₀Further, the maximum Lyapunov index corresponding to the bias angle phi, theta, psi can be obtained as shown in the following table.

TABLE 1 maximum Lyapunov index values of various state variables of a wind turbine tower

Because the maximum Lyapunov index values of all state variable time sequences are all larger than 0, the online monitoring system for the wind turbine tower attitude is a chaotic system, and the time sequences need to be analyzed and processed by using a correlation method of a chaos theory.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

As shown in fig. 1, a wind turbine tower attitude prediction method includes the following steps:

step 101: and acquiring the multivariate time sequence data of the wind turbine tower attitude. The multivariate time series data comprises angular rates of the wind turbine tower along an x-axis, a y-axis, and a z-axis, and tilt angles of the wind turbine tower along the x-axis, the y-axis, and the z-axis.

Step 102: and performing phase space reconstruction on the multi-element time sequence data by using an improved C-C method. The method specifically comprises the following steps: determining delay time of each variable in the multivariate time sequence data; calculating an embedding dimension; performing a phase space reconstruction of the multivariate time series data based on the delay time and the embedding dimension.

Considering a multivariable discrete time sequence of the wind turbine tower attitude online monitoring system:

X(t_p)＝[x₁(t_p),…,x_n(t_p)]^T

wherein time t_p1,2, …, N being the number of data in time series, x₁(t_p),…,x_n(t_p) N states of the corresponding system at t_pThe value of the time of day. The multivariate time series data needs to be normalized before phase space reconstruction.

(1) After the multivariate time sequence is normalized, respectively aiming at each variable, selecting proper delay time tau_dThe final embedding dimension m is calculated. The following 2 quantities were calculated:

wherein the content of the first and second substances,in order to assist in calculating the value 1,to assist in calculating quantity 2. r is_iI σ/2 is the search radius taken in the calculation, and σ is the standard deviation of the time series. ByT to calculate τ corresponding to the first local minimum of_d＝tτ_sT is the time lag, τ_sIs the sampling interval. The maximum and minimum two search radii are chosen, defining the delta Δ S (m, t) as:

ΔS(m,t)＝max{S(m,r_i,t)}-min{S(m,r_i,t)}

wherein S (m, r, t) is defined as:

wherein m is 2,3, …; the correlation integral of the embedded time series is:

since the length N of the actual time series cannot be infinite, the calculation is specifically performed

As an estimate of C (m, r, t). Wherein, M ═ N- (M-1) t is the embedding point number of the M-dimensional space, Θ (·) is the Heaviside function, Θ (x) is 1, x is not less than 0, Θ (x) is 0, and x is less than 0; i | · | purple wind_∞Is 1_∞And (4) norm. At the same time with

As an estimate of S (m, r, t). Thus, the phase space reconstruction is performed on each variable to obtain a multivariate time series embedded delay vector as follows:

(2) the conventional C-C process in step 1 is modified as follows:

(1) replacing the original infinity norm with the 2-norm;

(2) σ ═ std (x) (1+ cv/3), cv ═ std (x)/mean (x) is defined so as to enlarge the search radius r;

(3) defining r ═ rlog (m +1) to enlarge the search radius and reduce oscillation, the correlation integral of the improved embedding time sequence is now:

wherein r (m) > 0, and when m ═ 1, r (m) > r; when m > 1, r (m) rlog (m + 1). R in other formulas is also changed into r (m);

(4) by using the idea of 'averaging' in spectrum estimation, the time series is divided into multiple segments, and tau of each segment of the time series is obtained_dAnd τ_wTake τ_dIs taken as the final delay time and the final embedding dimension m is calculated. Solving for τ for each variable_dAnd m is followed by takingm^*Max (m) as the final embedded delay parameter for the multivariate time series. Its embedded delay vector is as follows:

step 103: constructing a plurality of corresponding initial prediction models for the multivariate time sequence data after the phase space reconstruction; the initial prediction model is a support vector regression model.

Step 104: and optimizing each initial prediction model by adopting a sequential minimum method.

The principle of step 103-104 is as follows:

(1) standard support vector machine (SVR) method

1) Mapping the inclination angle and the angular rate measured by the on-line wind turbine tower attitude monitoring system to a high-dimensional space F and constructing a regression estimation function F (x) in the space to define as follows:

f(x)＝＜w·Φ(x)＞+b

wherein the dimension of w is the feature space dimension, phi represents the nonlinear mapping from the input space to the output space, and b is the threshold. Solving for w and b by minimizing the objective function, the formula is:

wherein epsilon is a loss function and is used for controlling the size of the regression approximation error pipeline; c is a penalty factor used for balancing the flatness degree and the number of sample points with the deviation larger than epsilon of the regression function f; xi_iAndis a relaxation factor.

2) The regression function is very sensitive to isolated points due to the fixed penalty factor C, which in turn results in errors. The invention adopts a linear distribution method to obtain a multi-element time sequence weighting coefficient rho after phase space reconstruction, and the formula is as follows:

1≤i≤nm^*

wherein eta is the initial degree of relationship, and n is the variable number of the time series. Thus, the rewritable optimization objective function is:

introducing a Lagrange multiplier, and constructing the following Lagrange function:

for parameters w, b, xi_i,Calculating the partial derivatives, and making the partial derivatives of all parameters zero, the following dual optimization problem can be obtained:

(i＝1,2,…,l)

at this time, the function regression problem of the support vector machine can be summarized as a quadratic programming problem, and the regression estimation function is as follows:

wherein, K (x, x)_i) Is a kernel function. Because of alpha_iAndcannot be 0 at the same time, so b can be calculated as:

(2) the initial prediction model is optimized using Sequential Minimal Optimization (SMO).

According to the characteristics of the attitude monitoring data of the tower drum of the fanThe invention uses SMO-SVR method to respectively establish prediction models of 6 state quantities, and the algorithm adopts two layers of circulation optimization parameters alpha₁And alpha₂As shown in fig. 2.

The algorithm adopts two layers of circulation to optimize the parameter alpha₁And alpha₂. Outer layer circulation: for parameter alpha₂Selection is performed. And finding out sample points violating the KKT condition from the non-boundary points of the sample data, if the sample points do not exist, searching the whole sample set, and if the sample points do not exist, finishing the algorithm. Inner layer circulation: finding out | E (alpha) from non-boundary sample points₁)-E(α₂) The optimum parameter α of | max₁Where E (-) is an error function. If such an alpha is not present₁Then alpha is selected in the outer layer cycle₂And (5) carrying out outer circulation again for invalid parameters. The SMO algorithm process is described as follows:

1) for initially selected alpha₁And alpha₂To update these two values, the new values of the multipliers must be in a straight line in order not to violate the linear constraint, as follows:

in the formula (I), the compound is shown in the specification,is alpha₁And alpha₂The average value of the corresponding weight coefficient. Thus, the argument of the objective function is limited to the plane (α)₁,α₂) The two-dimensional optimization problem becomes a one-dimensional problem, so that the solution can be analyzed.

2) First, solve forThen it is used to solve Is operable asWherein:

when y is₁≠y₂,

When y is₁＝y₂,

3) By solving for alpha₂The optimized values of (A) are:

then, α can be solved₁Let s be y₁y₂And then:

4) correcting the threshold b:

wherein the content of the first and second substances,are respectively asAndexpressions when not on a boundary. If it isAndare not on the boundary, thenIf it isAndall on the boundary, taking the average value of the two formulas,

5) the shutdown criteria for the KKT condition are improved. If (w, xi)^*) Is the solution of the original problem of the objective function, and is simultaneously (alpha )^*) For the solution of the dual problem, then the corresponding objective function values are made equal, i.e. R (w, xi)^*)＝W(α,α^*) From this, it is possible to obtain:

order toThen:

for a given accuracy requirement, whenBelow this precision, the algorithm terminates. HandleThe shutdown criteria are used in conjunction with the shutdown criteria for the KKT condition, which may be set to occur once after a number of iterationsChecking to improve the running efficiency of the algorithm.

Step 105: and training each optimized prediction model based on the multi-element time sequence data after the phase space reconstruction. The method specifically comprises the following steps: and inputting the multivariate time sequence data after the phase space reconstruction into each optimized prediction model to obtain output data, and optimizing penalty factors and nuclear parameters in each optimized prediction model based on the output data to finish the training process.

According to the measured state parameter characteristics of the wind turbine tower, the invention selects a radial basis function K (x)_i,x_j)＝exp(-β||x_i-x_j||²) As a kernel function, β represents a kernel parameter. At the moment, the penalty factor C and the nuclear parameter beta need to be optimized according to the actual collected data of the state quantity of the wind turbine tower, so that the best prediction performance is obtained.

Setting C to be the range of output sample data prevents C from being overly sensitive to singular values, in the form:

wherein the content of the first and second substances,representing the mean value, En, of the output samples of the wind turbine tower_yAnd outputting the entropy of the sample for the wind turbine tower calculated by utilizing the inverse cloud. Since β is the expansion coefficient of the radial basis kernel, and the expansion coefficient generally reflects the range of the wind turbine tower input samples, it can be said that:

in formula (II), En'_xFor En of the input samples from the tower of a wind turbine_xAndthe normally-distributed random number is calculated,

step 106: and predicting the attitude of the tower drum of the wind turbine through the trained prediction models.

And performing one-step prediction by using the trained prediction model, bringing a new prediction value into the initially acquired state quantity time sequence, removing the most original sample data, then training the new prediction model again, and repeating iterative prediction.

Compared with the prior art, the invention has the beneficial effects that:

(1) in order to solve the problems of nonlinearity and chaos existing in time sequence data, a quantitative time sequence prediction method based on machine learning is provided. The method is characterized in that deep research is carried out on a representative support vector regression method, a phase space reconstruction method is utilized to process a time sequence, a weighted SVR prediction model is established, and meanwhile, the shutdown criterion and model parameters of the model are optimized;

(2) in order to make up for the respective defects of the qualitative and quantitative prediction methods, the two methods are organically combined, and a time sequence prediction method based on qualitative/quantitative mixing is provided to solve the problems of nonlinearity, chaos and uncertainty at the same time, and mainly comprises a cloud network method. Furthermore, a machine learning method is used for replacing matrix calculation, a multi-factor high-order cloud network hybrid prediction model can be established, and the prediction performance of the hybrid model can be further improved.

The specific embodiment is as follows:

firstly, the normalized variable time sequence to be predicted is subjected to phase space reconstruction by utilizing an improved C-C average method, and the embedding dimension is obtained asA delay time ofAnd optimizing the penalty factor C and the nuclear parameter gamma aiming at input and output samples consisting of 6 variables to be predicted and auxiliary prediction factors thereof, wherein fitness curves are shown in figures 3 to 8.

The optimum parameters obtained are shown in table 1:

TABLE 1 optimal parameters table

Finally, the prediction model is also trained by using 60 sets of time series, and fig. 9 to 14 show the one-step prediction simulation results of 22s to 32s from the beginning of the fault.

As can be seen from fig. 9 to 14, the SVR model has better short-term prediction capability, and the prediction accuracy thereof is closely related to the reliability of the training samples and the determination of the initial parameters.

Further, as with the cloud fuzzy method, a main factor domain U and k auxiliary predictor domains V are defined^k. Still taking the variable phi as an example, the domain of argument for the main factor phi is defined as U [ -0.0065,0.0065]The argument field of the auxiliary predictor r is V [ -0.0030,0.0025]And dividing U and V into 10 domain subintervals, and setting parameters C and gamma, such as C, of SVR1 and SVR2₁＝2.43,C₂＝19,γ₁＝0.086,γ₂0.0014. According to the above process, the simulation results of one-step prediction from 22s to 32s for each variable to be predicted are shown in fig. 15 to 20.

As can be seen from the simulation results of the state variables, the cloud-SVR time series prediction method has a very good effect on short-term prediction, and can accurately track various fluctuation changes of the time series data.

The invention also provides a wind turbine generator tower attitude prediction system, which comprises:

and the multivariate time sequence data acquisition module is used for acquiring multivariate time sequence data of the wind turbine tower attitude.

And the phase space reconstruction module is used for performing phase space reconstruction on the multi-element time sequence data by utilizing an improved C-C method.

The model construction module is used for constructing a plurality of corresponding initial prediction models for the multivariate time series data after the phase space reconstruction; the initial prediction model is a support vector regression model.

And the optimization module is used for optimizing each initial prediction model by adopting a sequential minimum method.

And the training module is used for training each optimized prediction model based on the multi-element time sequence data after the phase space reconstruction.

And the prediction module is used for predicting the attitude of the tower drum of the wind turbine through the trained prediction models.

Wherein, the training module specifically includes:

the input unit is used for inputting the multivariate time sequence data after the phase space reconstruction into each optimized prediction model to obtain output data;

and the optimization unit is used for optimizing the penalty factors and the nuclear parameters in each optimized prediction model based on the output data to complete the training process.

Wherein, the phase space reconstruction module specifically comprises:

a delay time determining unit for determining a delay time of each variable in the multivariate time series data;

an embedding dimension calculation unit for calculating an embedding dimension;

and the phase space reconstruction unit is used for performing phase space reconstruction on the multivariate time sequence data based on the delay time and the embedding dimension.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.