A thermal monitoring parameter noise reduction method and system based on modal double decomposition
Through modal double decomposition technology and chaotic time series analysis, the problem of difficulty in removing noise and disturbances in thermal monitoring signals is solved, and the signal quality improvement and the accuracy of equipment health assessment are achieved.
Patent Information
- Application Number
- CN202410192348.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-21
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-02-21
AI Technical Summary
The prior art is difficult to effectively remove noise and disturbances in thermal monitoring signals, resulting in a decrease in signal quality, affecting the accuracy of the integrated control system and the evaluation of equipment health status.
Using a modal double decomposition method, MREMD and VMD decomposition technology combined with the arrangement entropy value of chaotic time series analysis, the noise part is screened and reconstructed to maximize the preservation of the original signal information and reduce noise.
A higher signal-to-noise ratio and lower information entropy and power spectrum entropy are achieved, which significantly improves signal quality and enhances comprehensive understanding of device status and health assessment capabilities.
Smart Images

Figure CN117972327B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data denoising, and in particular to a thermal monitoring parameter denoising method based on modal double decomposition and a thermal monitoring parameter denoising system based on modal double decomposition. Background Art
[0002] The thermal system is an organic combination of a series of high-temperature, high-pressure and high-speed equipment. There are various forms of connections between the various components of the system, such as mechanical, pneumatic, thermal, heat exchange, and mass flow. Any change in the state of any device or process will have a multi-faceted impact on the system. In order to ensure the operational stability of the system, the industry generally uses an integrated control system to control, optimize and analyze its operating process. In this process, the signal quality obtained by the monitoring system and the amount of information contained in the signal will directly affect the work of the integrated control system, and then determine the operation and management level of the entire system. Limited by the sampling frequency, accuracy and external environmental interference of the sensor, the sampling data will contain certain noise and disturbances. If they cannot be accurately removed, they may blur and cover up the real monitoring signal, mislead the analysis and decision of the integrated control system, and lead to misjudgment or omission of abnormal events; at the same time, noise and disturbances may also cause the loss or deformation of important information in the monitoring data, making it difficult to fully understand the system status, and thus unable to accurately evaluate the health status and performance of the equipment.
[0003] Therefore, it is necessary to determine the composition distribution of thermal monitoring signals based on an in-depth analysis of the characteristics of thermal monitoring parameter signals and the laws of spectrum distribution, and use the modal decomposition method to divide the original signal into an information part and a noise part. Then, the effective information in the noise part is screened out by performing secondary modal decomposition on the noise part, so as to achieve the maximum removal of noise and disturbance while retaining the original signal information as much as possible.
[0004] According to the publicly available literature, there is no noise reduction method for thermal monitoring data at home and abroad. The main reason is that thermal monitoring parameters are nonlinear, delayed, highly coupled, and highly disturbed. At the same time, the periodic fluctuations and oscillation frequencies of thermal signals also contain certain effective information. This information is mixed with noise in the intermediate frequency domain and is difficult to accurately separate.
[0005] At present, the noise reduction methods for thermal system monitoring parameters are mainly divided into improved wavelet noise reduction methods and improved traditional filtering methods. However, these methods have the disadvantages of blurred boundary effects, serious loss of details, and can only reduce noise but not remove it. Moreover, the above noise reduction methods are basically universal monitoring data noise reduction methods, which do not combine the characteristics of thermal monitoring parameters, resulting in more data information loss or incomplete noise reduction. Summary of the invention
[0006] In view of the above problems, the present invention provides a thermal monitoring parameter denoising method and system based on modal double decomposition, which obtains a number of intrinsic mode functions by performing MREMD (Median Regression Empirical Mode Decomposition) decomposition on the monitoring parameters, and then screens out the IMF (Intrinsic Mode Functions) components containing noise and reconstructs them into the noise part of the original data by introducing the permutation entropy value of chaotic time series analysis, and finally extracts the original data information in the noise part through VMD (Variational Mode Decomposition) decomposition, and reconstructs them into denoised data together with the residual IMF components containing trends and signals obtained by MREMD decomposition, so as to reduce the noise while retaining the information amount of the original signal to the maximum extent, and has a higher signal-to-noise ratio and lower information entropy and power spectrum entropy.
[0007] To achieve the above object, the present invention provides a thermal monitoring parameter denoising method based on modal double decomposition, comprising:
[0008] Perform MREMD decomposition on the operating parameters of the thermal system to obtain the IMF component and residual component;
[0009] Calculating the permutation entropy value of the IMF component based on the CC (cross-correlation) algorithm, performing K-means clustering on the permutation entropy value to obtain a high-frequency IMF component and a low-frequency IMF component respectively;
[0010] summing and reconstructing the high-frequency IMF components to obtain the noise components of the operating parameters of the thermal system;
[0011] The VMD decomposition parameters are obtained by optimizing the NGO algorithm of the northern goshawk, and the VMD decomposition is performed on the noise component to obtain the IMF* component;
[0012] The IMF* component with the lowest envelope entropy is used to reconstruct the residual component and the low-frequency IMF component to obtain the original data after noise reduction.
[0013] In the above technical solution, preferably, the specific method for calculating the permutation entropy value of the IMF component based on the CC algorithm includes:
[0014] The permutation entropy value is calculated for the IMF component, and the CC algorithm in chaotic system analysis is used to iteratively calculate and optimize the time delay and embedding dimension of the phase space reconstruction of the IMF component to obtain the permutation entropy value of the IMF component.
[0015] In the above technical solution, preferably, the VMD decomposition parameters are obtained by optimizing the NGO algorithm by the northern goshawk, and the VMD decomposition is performed on the noise component to obtain the IMF* component, and the specific method includes:
[0016] The IMF components are decomposed and extracted using a variational mode decomposition (VMD) algorithm, and the obtained IMF components are recorded as IMF* components;
[0017] The penalty factor and the number of decomposition layers of the VMD algorithm are optimized by using the NGO algorithm to obtain the optimal solution of the penalty factor and the number of decomposition layers of the VMD algorithm within a preset iteration range;
[0018] The IMF* component is determined under the penalty factor and the number of decomposition levels of the optimal solution.
[0019] In the above technical solution, preferably, the specific process of reconstructing using the IMF* component with the lowest envelope entropy, the residual component and the low-frequency IMF component includes:
[0020] Calculating the envelope entropy of the IMF* component, and using the envelope entropy as the fitness of the NGO algorithm in the VMD decomposition parameter optimization process;
[0021] When the envelope entropy of the IMF* component is the lowest, the fitness of the NGO algorithm is optimal, and the IMF* component at this time is used to reconstruct the residual component and the low-frequency component.
[0022] In the above technical solution, preferably, the thermal monitoring parameter denoising method based on modal dual decomposition further includes:
[0023] According to the sliding window method, the operating parameters of the thermal system are denoised within a preset window.
[0024] The present invention further proposes a thermal monitoring parameter denoising system based on modal double decomposition, which uses the thermal monitoring parameter denoising method based on modal double decomposition disclosed in any one of the above technical solutions, including:
[0025] MREMD decomposition module, used to perform MREMD decomposition on the operating parameters of the thermal system to obtain IMF components and residual components;
[0026] An IMF component clustering module is used to calculate the permutation entropy value of the IMF component based on the CC algorithm, and perform K-means clustering on the permutation entropy value to obtain a high-frequency IMF component and a low-frequency IMF component respectively;
[0027] A high-frequency noise reconstruction module, used for summing and reconstructing the high-frequency IMF components to obtain the noise components of the thermal system operation parameters;
[0028] A VMD component decomposition module, used for obtaining VMD decomposition parameters by optimizing the NGO algorithm of the northern goshawk, and performing VMD decomposition on the noise component to obtain an IMF* component;
[0029] The component reconstruction and denoising module is used to reconstruct the IMF* component with the lowest envelope entropy with the residual component and the low-frequency IMF component to obtain the original data after denoising.
[0030] In the above technical solution, preferably, the IMF component clustering module is specifically used for:
[0031] The permutation entropy value is calculated for the IMF component, and the CC algorithm in chaotic system analysis is used to iteratively calculate and optimize the time delay and embedding dimension of the phase space reconstruction of the IMF component to obtain the permutation entropy value of the IMF component.
[0032] In the above technical solution, preferably, the VMD component decomposition module is specifically used for:
[0033] The IMF components are decomposed and extracted using a variational mode decomposition (VMD) algorithm, and the obtained IMF components are recorded as IMF* components;
[0034] The penalty factor and the number of decomposition layers of the VMD algorithm are optimized by using the NGO algorithm to obtain the optimal solution of the penalty factor and the number of decomposition layers of the VMD algorithm within a preset iteration range;
[0035] The IMF* component is determined under the penalty factor and the number of decomposition levels of the optimal solution.
[0036] In the above technical solution, preferably, the component reconstruction denoising module is specifically used for:
[0037] Calculating the envelope entropy of the IMF* component, and using the envelope entropy as the fitness of the NGO algorithm in the VMD decomposition parameter optimization process;
[0038] When the envelope entropy of the IMF* component is the lowest, the fitness of the NGO algorithm is optimal, and the IMF* component at this time is used to reconstruct the residual component and the low-frequency component.
[0039] In the above technical solution, preferably, the thermal monitoring parameter denoising system based on modal dual decomposition also includes a sliding window denoising module, which is used to denoise the thermal system operating parameters within a preset window according to a sliding window method.
[0040] Compared with the prior art, the beneficial effects of the present invention are as follows: a number of intrinsic mode functions are obtained by performing MREMD decomposition on the monitoring parameters, and then the IMF components containing noise are screened out and reconstructed into the noise part of the original data by introducing the permutation entropy value of chaotic time series analysis, and finally the original data information in the noise part is extracted through VMD decomposition, and reconstructed into denoised data together with the residual IMF components containing trends and signals obtained by MREMD decomposition, which reduces the noise while retaining the information amount of the original signal to the maximum extent, and has a higher signal-to-noise ratio and lower information entropy and power spectrum entropy. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 A schematic flow chart of a thermal monitoring parameter denoising method based on modal double decomposition disclosed in an embodiment of the present invention;
[0042] Figure 2 A schematic diagram of a module of a thermal monitoring parameter noise reduction system based on modal dual decomposition disclosed in an embodiment of the present invention;
[0043] Figure 3 A schematic diagram of a pure signal of a boiler drum water level change obtained by simulating a steam power generation system disclosed in an embodiment of the present invention;
[0044] Figure 4 A schematic diagram of adding white noise to a pure signal in a steam power generation system disclosed in an embodiment of the present invention;
[0045] Figure 5 A schematic diagram of obtaining components by performing MREMD decomposition on a simulated signal containing noise according to an embodiment of the present invention;
[0046] Figures 6 to 8 A schematic diagram of the statistical iteration process of each IMF component disclosed in an embodiment of the present invention;
[0047] Fig. 9 A schematic diagram of summing and reconstructing high-frequency IMF components into the noise part of a simulation signal disclosed in an embodiment of the present invention;
[0048] Fig.10 A schematic diagram of the fitness change of the NGO optimization process disclosed in an embodiment of the present invention;
[0049] Fig.11 A schematic diagram of IMF* components obtained by VMD decomposition disclosed in an embodiment of the invention;
[0050] Fig.12 A schematic diagram of a comparison between a noise-reduced signal, a pure signal and a noisy signal disclosed in an embodiment of the invention;
[0051] Fig.13A schematic diagram of comparing a noise-reduced signal with an original signal disclosed in an embodiment of the invention.
[0052] In the figure, the corresponding relationship between each component and the reference numeral is as follows:
[0053] 1. MREMD decomposition module, 2. IMF component clustering module, 3. High-frequency noise reconstruction module, 4. VMD component decomposition module, 5. Component reconstruction denoising module, 6. Sliding window denoising module. DETAILED DESCRIPTION
[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0055] The present invention is further described in detail below in conjunction with the accompanying drawings:
[0056] like Figure 1 As shown, a thermal monitoring parameter denoising method based on modal double decomposition provided by the present invention comprises:
[0057] Perform MREMD decomposition on the operating parameters of the thermal system to obtain the IMF component and residual component;
[0058] The permutation entropy values of IMF components are calculated based on CC algorithm, and K-means clustering is performed on the permutation entropy values to obtain high-frequency IMF components and low-frequency IMF components respectively.
[0059] The high-frequency IMF components are summed and reconstructed to obtain the noise components of the thermal system operating parameters;
[0060] The VMD decomposition parameters are obtained by optimizing the NGO algorithm with the Northern Goshawk, and the noise component is decomposed by VMD to obtain the IMF* component;
[0061] The IMF* component with the lowest envelope entropy is reconstructed with the residual component and the low-frequency IMF component to obtain the original data after denoising.
[0062] In this embodiment, several intrinsic mode functions are obtained by performing MREMD decomposition on the monitoring parameters, and then the IMF components containing noise are screened out and reconstructed into the noise part of the original data by introducing the permutation entropy value of chaotic time series analysis. Finally, the original data information in the noise part is extracted through VMD decomposition, and reconstructed into denoised data together with the residual IMF components containing trends and signals obtained by MREMD decomposition. While reducing the noise, the information amount of the original signal is retained to the maximum extent, with a higher signal-to-noise ratio and lower information entropy and power spectrum entropy.
[0063] Specifically, the implementation steps of this method are divided into the following four steps:
[0064] Step 1: Perform MREMD decomposition on the operating parameters of the thermal system to obtain several IMF components and residual components;
[0065] Step 2: Calculate the permutation entropy value of each IMF component after optimizing the time delay and reconstruction dimension by CC algorithm, perform K-means clustering on the permutation entropy value of each IMF component, divide the IMF component into high-frequency IMF components containing noise and disturbance and low-frequency IMF components containing trend information, and reconstruct the high-frequency IMF components by summing and reconstructing them to obtain the noise component of the original data;
[0066] Step 3, using VMD decomposition to extract the information part of the noise component of the original data obtained in step 2, the VMD decomposition parameters are obtained by the Northern Goshawk optimized NGO algorithm, and the original data noise part is decomposed into several IMF* components;
[0067] Step 4: Record the fitness of the IMF* component at the optimal fitness in the NGO algorithm, select the IMF* component with the best fitness and reconstruct it with the low-frequency IMF component in step 2 and the residual component in step 1 as the original data after denoising.
[0068] Among them, MREMD is based on empirical mode decomposition, uses the autoregressive (AR) model to extend the signal endpoints, and uses the method of optimizing envelope fitting to improve the problem that EMD does not quantitatively distinguish between effective components and interference components, which easily causes prediction errors; the specific steps are as follows:
[0069] Step 1: Assume that the time series under study is s0(t), and use the AR model to extend both ends of s(t) until the left and right endpoints of the original sequence are between the two adjacent extreme points of the extended time series.
[0070] s t =φ0+φ1s t-1 +φ2s t-2 +…+φ p s t-p +μ t (1)
[0071] Where: s is the mean point; φ0, φ1, …φ p is p+1 real numbers; μ t (p+1,p+2,…,N) is a white noise sequence with zero mean.
[0072] Step 2: Assume that there are k extreme points in the extended s0(t), and find the average of the adjacent extreme points to get the sequence {s mi ,0}, where i = 1, 2, ..., k-1, and then the mean sequence is interpolated by cubic spline:
[0073] h 1,0 (t) = s0(t) - m 1,0 (t) (2)
[0074] Where: m 1,0 (t),s0(t)-m 1,0 (t) is the signal mean sequence of s0(t), h 1,0 (t) is the first-order signal component of s0(t).
[0075] Step 3: h 1,0 (t) Repeat steps 1 and 2 for iterative calculation. Suppose that after l iterations h 1,l (t) If the following termination conditions are met, the iteration stops at this time. The termination conditions are:
[0076]
[0077] Where: * and σ i are h after the l-1th iteration respectively. 1,t-1(t) The standard deviation of the mean point sequence; σ i After the lth iteration h 1,l (t) is the standard deviation of the mean point sequence; s z is the zth extreme point of the extended sequence; P is the conditional probability; θ0 and θ l is the ratio of the initial signal and the signal mean point after the lth iteration to xMσ, as shown in formula (4):
[0078]
[0079] Where: sm 1,0, sm 2,0, sm 3,0 are the mean points of the initial signal s0(t) before iteration; sm1,l, sm2,l, smk,l are the signal components h after the l-1th iteration 1,l-1 (t) is the mean point.
[0080] Step 4: h 1,l(t) is the first-order IMF component, x0(t)-IMF1 is the first-order residual signal R1, and R1 is used as the original signal to repeat steps 1 to 3 until the residual component becomes a monotonic function or no new IMF component can be separated. The calculation process is shown in formula (5):
[0081]
[0082] Where: n is the maximum number of IMF components that can be decomposed; Rn is the n-order residual signal of the original signal s0(t), called the residual component
[0083] After the above decomposition, the original signal s0(t) can be expressed as the sum of all IMF components and residual components, expressed as:
[0084]
[0085] On this basis, the IMF components obtained after the original signal is decomposed by MREMD can be divided into high-frequency IMF components containing noise and disturbance, low-frequency IMF components containing trend information, and intermediate frequency components containing both noise and trend information. Regarding the selection of the division standard of IMF components, since the IMF components are mainly different in the spectrum, the dimension gap between them and the original signal is large, and the evaluation standards based on dimension such as root mean square error cannot effectively evaluate the quality of IMF components. The present invention selects permutation entropy as the standard for evaluating the quality of IMF component signals and divides IMF components of each order.
[0086] As an indicator for quantitatively evaluating the randomness of time series, permutation entropy has the advantages of fast calculation speed, strong noise resistance, and no dependence on data length compared with traditional approximate entropy and sample entropy. It is often used to measure the complexity of time series.
[0087] The specific algorithm is shown in formula (7) to formula (10):
[0088] Step 1: Assume that a time series is y(k), and reconstruct the original data in phase space:
[0089]
[0090] Where: m is the embedding dimension, τ is the delay time, and K is the number of reconstructed components
[0091] Step 2: Rearrange each reconstructed component in the order of the first element to obtain the column of each element position in the vector, and the index constitutes a set of symbol sequences S(l)
[0092] S(l)={j1,j2,…,j m}; l = 1, 2, ..., k, and k ≤ m! (8)
[0093] Step 3: Calculate the probability of each symbol sequence appearing.
[0094]
[0095] Step 4: H pe Normalization is performed to make it more intuitive and convenient in practical applications:
[0096]
[0097] Where: f j is the probability of the symbol sequence appearing, H pe is the permutation entropy value.
[0098] In the above implementation, preferably, the specific method for calculating the permutation entropy value of the IMF component based on the CC algorithm includes:
[0099] The permutation entropy value of the IMF component is calculated. At the same time, the CC algorithm in chaotic system analysis is used to iteratively calculate and optimize the time delay and embedding dimension of the IMF component phase space reconstruction to obtain the permutation entropy value of the IMF component.
[0100] In this embodiment, since the thermal monitoring parameters are nonlinear and unstable time series, the IMF components obtained by MREMD decomposition have chaotic time series characteristics. Therefore, when calculating the permutation entropy of each IMF component, the present invention uses the CC method in chaotic system analysis to calculate the time delay τ of phase space reconstruction. b and embedding dimension m. The CC algorithm uses a cross-correlation function to measure the degree of correlation between the reconstructed data and the original data, and adjusts the phase space reconstruction parameters according to the degree of correlation. Through continuous iteration and parameter adjustment, the CC algorithm can gradually optimize the phase space reconstruction results, reduce information loss during dimensionality reduction and reconstruction, and retain the characteristics of the original data as much as possible, thereby improving the representation accuracy and interpretability of the phase space reconstruction data. The calculation process is as follows:
[0101] The IMF component time series {s M (i)}=s M (1), s M (2), …, s M (K) is decomposed into τ subsequences:
[0102]
[0103] Then calculate τ subsequences:
[0104]
[0105] When N→∞, we have:
[0106]
[0107] Since some elements in the IMF component have certain correlation, S(m,r,τ)≠0, and its maximum deviation is:
[0108]
[0109] According to the data length K, the standard deviation σ of the IMF component time series is calculated, and then the three statistics are calculated using formula (15)
[0110]
[0111] Take S in the iteration process cor The minimum value of (τ) is the optimal time delay value for phase space reconstruction, given by S cor The correlation between (τ) and τ is used to obtain the input dimension m of phase space reconstruction using formula (16):
[0112]
[0113] Where: τ d for The time delay value corresponding to the first minimum value.
[0114] According to the permutation entropy value of each order IMF component, the k-means clustering analysis algorithm is used to divide each IMF component into two categories according to the entropy value, which are the noise part and information part of the original data respectively. The IMF components with low permutation entropy are selected and reconstructed into the high-frequency noise component of the original data, and VMD decomposition and denoising are performed on this part.
[0115] In the above implementation, preferably, the VMD decomposition parameters are obtained by optimizing the NGO algorithm using the Northern Goshawk, and the noise component is subjected to VMD decomposition to obtain the IMF* component. The specific method includes:
[0116] The variational mode decomposition (VMD) algorithm is used to decompose and extract the IMF components, and the obtained IMF components are recorded as IMF* components;
[0117] The NGO algorithm is used to optimize the penalty factor and decomposition level of the VMD algorithm, and the optimal solution of the penalty factor and decomposition level of the VMD algorithm within the preset iteration range is obtained;
[0118] Determine the IMF* component under the penalty factor and number of decomposition levels of the optimal solution.
[0119] In this embodiment, for the high-frequency IMF components obtained by screening, the traditional method is to directly discard them to achieve the purpose of noise reduction. However, since the high-frequency IMF may contain original data information such as periodic fluctuations in addition to noise and disturbance, direct discarding will cause information loss. In response to this problem, the present invention adopts a method of performing modal decomposition again on the high-frequency IMF components obtained by MREMD decomposition of the original data, screening out the information part in the high-frequency IMF components, and reconstructing them with the low-frequency components and residual components, which can remove noise and irregular disturbances while retaining the original data information as much as possible.
[0120] VMD is a completely non-recursive variational mode decomposition. The overall idea is: assume that any signal is composed of K columns of intrinsic mode functions with specific center frequencies and finite bandwidths. Based on classical Wiener filtering, it is subject to the following two constraints: (1) the sum of the bandwidths of the center frequencies of each modal component is required to be minimum; (2) the sum of all modal components is equal to the original signal. Solve the variational problem, find the effective components corresponding to each center frequency in the frequency domain, and obtain the intrinsic mode function. In order to distinguish it from the IMF components obtained by MREMD decomposition, the IMF components obtained by VMD decomposition are recorded as IMF*. The specific steps are as follows:
[0121] Step 1: Let k IMF components be V k(t) , the center frequency is ω(t); V is obtained by Hilbert transform k(t) The analytical signal and its one-sided spectrum are calculated by using the operator e -jωkt Multiply V k(t) The center band of is modulated to the corresponding baseband, as shown in equation (17):
[0122]
[0123] Step 2: Calculate the square norm L of the demodulated gradient 2 , and estimate the bandwidth of each mode component. The process is shown in formula (18):
[0124]
[0125] Where: k ﹜=﹛v1,…,v k ﹜ represents the decomposed IMF component; ﹛ω k ﹜ is the center frequency of each component;
[0126] In order to find the optimal solution to the constrained variational problem, the extended Lagrangian expression shown in Equation 3.3.3 is used:
[0127]
[0128] Where: τ(t) is the Lagrange multiplier; α is the second-order penalty factor.
[0129] Step 3: Use the alternating direction multiplier method (ADMM) to continuously update each component and its center frequency, and finally obtain the saddle point of the unconstrained model, which is the optimal solution of the original model. All components can be obtained according to the frequency domain space by formula (19):
[0130]
[0131] Where: ω is the frequency; are the Fourier transforms of vn+1k(t), s(t), and τ(t) respectively;
[0132] Step 4: vn+1k(ω) is The residual after Wiener filtering is determined by whether the power spectrum center of each component meets the constraint conditions. If yes, the IMF component is output, otherwise it continues to iterate. The specific process is:
[0133] initialization and n
[0134] (1) n = n + 1, enter the loop
[0135] Update k and The inner loop stops when the decomposition level K is reached.
[0136]
[0137]
[0138] (4) Repeat (2) to (3) until the condition shown in formula 3.3.7 is met:
[0139]
[0140] In VMD decomposition, the penalty factor α and the number of decomposition levels K need to be determined by themselves, and the selection of their values will directly affect the decomposition effect of the VMD method. In view of this problem, the present invention proposes to use the Northern Goshawk algorithm to find the optimal solution of K and α within a certain iteration range.
[0141] The Northern Goshawk Algorithm is a global optimization algorithm proposed by Mohammad Dehghani et al. in 2022, inspired by the hunting behavior of the Northern Goshawk. The algorithm has a novel structure, few parameters, and fast convergence. The specific optimization process is divided into prey identification and attack (exploration phase), pursuit and escape (exploitation phase). The specific algorithm steps are as follows:
[0142] First, the goshawk population is initialized, as shown in formula (23):
[0143]
[0144] Where: X is the population matrix of goshawk; x ij is the j-th position of the ith goshawk; N is the population size; m is the dimension of the problem to be solved
[0145] The objective function value of the goshawk population can be expressed by the objective function value vector shown in formula (24):
[0146]
[0147] Where: F is the objective function vector of the northern goshawk population; F i is the objective function value of the i-th northern goshawk
[0148] The northern goshawk's choice of prey when hunting is random, which is the exploration ability of the NGO algorithm. This stage is a global search of the search space to determine the optimal area. This stage is described by equations (25) to (27):
[0149]
[0150]
[0151]
[0152] Where: P i is the position of the i-th prey; F Pi is the objective function value of the i-th prey position; k* is a random integer in the range of [1,N]; Xnew,P1 i is the new position of the i-th northern goshawk; Xnew,P1 i,j is the new position of the i-th northern goshawk in the jth dimension; Fnew,P1 i is the objective function value of the i-th northern goshawk after updating based on the first stage; r is a random number in the range of [0,1]; I is a random integer of 1 or 2.
[0153] After the northern goshawk attacks its prey, the prey will try to escape, and the northern goshawk needs to continue chasing the prey at high speed, which is the local search of the algorithm in the search space. Assuming that this hunting activity is close to an attack position with a radius of R, it is called the development stage of the attack range. The process of this stage is described by equations (28) to (30):
[0154]
[0155] R=0.02(1-t / T) (29)
[0156]
[0157] Wherein: t is the current iteration number; T is the maximum iteration number; Xnew,P2 i is the new position of the i-th northern goshawk; Xnew,P2 i,j is the new position of the j-th dimension of the i-th northern goshawk; Xp2 i,j is the j-th dimension position of the i-th northern goshawk after updating based on the second stage, and Fnew,P2 i is the objective function value of the i-th northern goshawk after updating based on the second stage.
[0158] In the above implementation, preferably, the specific process of reconstructing using the IMF* component with the lowest envelope entropy, the residual component and the low-frequency IMF component includes:
[0159] Calculate the envelope entropy of the IMF* component and use it as the fitness of the NGO algorithm for the VMD decomposition parameter optimization process;
[0160] When the envelope entropy of the IMF* component is the lowest, the fitness of the NGO algorithm is optimal, and the IMF* component at this time is used to reconstruct the residual component and the low-frequency component.
[0161] In this implementation, since the calculation process of the permutation entropy of the CC algorithm to optimize the phase space reconstruction parameters is relatively complicated, and the NGO algorithm needs to repeatedly iterate the fitness function during the optimization process, the permutation entropy of the optimized phase space reconstruction parameters as the NGO fitness function will cause the calculation time to be too long, resulting in a waste of computing power. Although it can be used as an evaluation and division standard for high- and low-frequency signal sequences, it is not suitable for the NGO to optimize the fitness function of VMD.
[0162] Envelope entropy can be used to measure signal complexity and uncertainty by measuring the energy distribution of the signal on the spectrum, and then quantitatively analyze the spectral characteristics and frequency domain information of the signal. The higher the value of envelope entropy, the more evenly the spectral energy distribution of the signal is, and the greater the dispersion of the frequency domain information is; conversely, the lower the value of envelope entropy, the more concentrated the spectral energy distribution of the signal is, and the higher the concentration of the frequency domain information is. Since this method is simple to calculate, the present invention selects the envelope entropy of each IMF* component as the fitness function of NGO optimization, and the calculation method is shown in formula (31)-formula (33):
[0163]
[0164]
[0165]
[0166] Where: E p is the envelope entropy of the IMF* component; l(j) represents the envelope signal sequence obtained after Hilbert demodulation of the signal x(j) (j = 1, 2, ..., m); H represents the Hilbert transform of the signal; l(j) is the normalized form of a(j).
[0167] In the above-mentioned embodiment, preferably, the thermal monitoring parameter denoising method based on modal dual decomposition further includes: denoising the thermal system operating parameters within a preset window according to the sliding window method. The thermal monitoring parameter denoising method based on modal dual decomposition can be combined with the sliding window method as a data processing method within a window. By adding a denoising module to the original thermal monitoring system, the purpose of real-time denoising of the monitoring parameters can be achieved, and the original human-computer interaction components of the thermal system can be fully utilized to provide high-quality data to be analyzed for the integrated control system, thereby improving the operation management level of the entire system.
[0168] like Figure 2 As shown, the present invention further proposes a thermal monitoring parameter denoising system based on modal double decomposition, which applies the thermal monitoring parameter denoising method based on modal double decomposition disclosed in any one of the above embodiments, including:
[0169] MREMD decomposition module 1, used for performing MREMD decomposition on the operating parameters of the thermal system to obtain IMF components and residual components;
[0170] IMF component clustering module 2 is used to calculate the permutation entropy value of the IMF component based on the CC algorithm, perform K-means clustering on the permutation entropy value, and obtain high-frequency IMF components and low-frequency IMF components respectively;
[0171] A high-frequency noise reconstruction module 3 is used to sum and reconstruct the high-frequency IMF components to obtain the noise components of the thermal system operation parameters;
[0172] VMD component decomposition module 4, used for obtaining VMD decomposition parameters by optimizing NGO algorithm with Northern Goshawk, and performing VMD decomposition on noise components to obtain IMF* components;
[0173] The component reconstruction and noise reduction module 5 is used to reconstruct the IMF* component with the lowest envelope entropy with the residual component and the low-frequency IMF component to obtain the original data after noise reduction.
[0174] In this embodiment, several intrinsic mode functions are obtained by performing MREMD decomposition on the monitoring parameters, and then the IMF components containing noise are screened out and reconstructed into the noise part of the original data by introducing the permutation entropy value of chaotic time series analysis. Finally, the original data information in the noise part is extracted through VMD decomposition, and reconstructed into denoised data together with the residual IMF components containing trends and signals obtained by MREMD decomposition. While reducing the noise, the information amount of the original signal is retained to the maximum extent, with a higher signal-to-noise ratio and lower information entropy and power spectrum entropy.
[0175] In the above implementation, preferably, the IMF component clustering module 2 is specifically used for:
[0176] The permutation entropy value of the IMF component is calculated. At the same time, the CC algorithm in chaotic system analysis is used to iteratively calculate and optimize the time delay and embedding dimension of the IMF component phase space reconstruction to obtain the permutation entropy value of the IMF component.
[0177] In the above implementation, preferably, the VMD component decomposition module 4 is specifically used for:
[0178] The variational mode decomposition (VMD) algorithm is used to decompose and extract the IMF components, and the obtained IMF components are recorded as IMF* components;
[0179] The NGO algorithm is used to optimize the penalty factor and decomposition level of the VMD algorithm, and the optimal solution of the penalty factor and decomposition level of the VMD algorithm within the preset iteration range is obtained;
[0180] Determine the IMF* component under the penalty factor and number of decomposition levels of the optimal solution.
[0181] In the above implementation, preferably, the component reconstruction denoising module 5 is specifically used for:
[0182] Calculate the envelope entropy of the IMF* component and use it as the fitness of the NGO algorithm for the VMD decomposition parameter optimization process;
[0183] When the envelope entropy of the IMF* component is the lowest, the fitness of the NGO algorithm is optimal, and the IMF* component at this time is used to reconstruct the residual component and the low-frequency component.
[0184] In the above embodiment, preferably, the thermal monitoring parameter denoising system based on modal dual decomposition further includes a sliding window denoising module 6 for denoising the thermal system operating parameters within a preset window according to a sliding window method.
[0185] According to the thermal monitoring parameter denoising system based on modal double decomposition disclosed in the above-mentioned embodiment, the functions to be implemented by each module thereof correspond to the respective steps in the thermal monitoring parameter denoising method based on modal double decomposition disclosed in the above-mentioned embodiment. During the implementation process, the operations are performed with reference to the above-mentioned embodiment, which will not be repeated here.
[0186] According to the thermal monitoring parameter denoising method and system based on modal dual decomposition disclosed in the above-mentioned embodiment, the effects of the method and system are described through the following examples.
[0187] Example 1
[0188] Taking a certain type of ship steam power generation system as an example, when the turbine feed water pump and the electric feed water pump are powered off, the operating condition of the main engine intake valve opening increasing from 50% to 80% is simulated. The pure signal of the boiler drum water level change obtained by simulation is as follows: Figure 3 As shown in , after adding white noise with an amplitude of 0.1 times the original signal to the pure signal, Figure 4 Taking the noisy boiler drum water level change data as an example, the thermal monitoring parameter noise reduction method proposed in the present invention is verified.
[0189] Perform MREMD decomposition on the noisy simulation signal to obtain 6 IMF components and 1 residual component, such as Figure 5 shown.
[0190] According to the method shown in equations (7) to (10), the permutation entropy of IMF1 to 6 components is calculated, and the phase space reconstruction parameters are obtained by the CC algorithm shown in equations (11) to (16). The statistics S(τ), ΔS(τ), and Scor(τ) of each IMF component are as follows: Figure 6-Figure 8 shown.
[0191] Take the time delay value corresponding to the minimum value of Scor(τ) in the figure as the optimal time delay value of each IMF component; then calculate the optimal embedding dimension of each IMF according to formula (16); then substitute it into formula (7) to formula (10) to calculate the permutation entropy value of each IMF component. The optimal delay time, optimal embedding dimension and permutation entropy value of each IMF component are shown in Table 1.
[0192] Table.1 The best delay time, best embedding dimension and permutation entropy value of each IMF component
[0193]
[0194] Perform k-means clustering on the permutation entropy, select the IMF1, IMF2, IMF3, and IMF5 components with high permutation entropy and sum them to reconstruct the noise part of the simulation signal, such as Fig. 9 shown.
[0195] The VMD decomposition of equations (17) to (22) is performed on the noise data. The structural parameters of the VMD decomposition are obtained by NGO. The NGO population is set to 20, the number of iterations is 15, and the fitness function is the envelope entropy shown in equations (31) to (33). The fitness changes as follows: Fig.10 shown.
[0196] According to the NGO global optimization results, the optimal penalty coefficient α of VMD decomposition is 3000, the optimal decomposition layer number K is 10, and the optimal fitness is 3.8264. The envelope entropy of each IMF* component at the optimal fitness is shown in Table 2. Fig.11 shown.
[0197] Table 2 Envelope entropy of each IMF* component at optimal fitness
[0198]
[0199] The IMF*1 component obtained by VMD decomposition and the low permutation entropy IMF4, IMF6 components and residual components obtained by MREMD decomposition are reconstructed into a denoised simulation signal. The comparison between the denoised signal and the pure signal and the noisy signal is as follows: Fig.12 shown.
[0200] The signal-to-noise ratio of the denoised signal and the root mean square error between the denoised signal and the pure signal are calculated according to formula (34) as the evaluation criteria for the denoising effect.
[0201]
[0202] Where: P S is the pure signal power, P N is the noise power, K is the data length, s i is the signal after noise reduction, and h0 is the pure signal.
[0203] After calculation, the signal-to-noise ratio of the signal after denoising is 34.1637; the root mean square error is 0.0119. Compared with the trend extraction based on MWSA, wavelet denoising with improved threshold, traditional soft threshold wavelet denoising and hard threshold wavelet denoising, the results are shown in Table 3.
[0204] Table 3 Comparison of denoising methods based on mode dual decomposition and other denoising methods
[0205]
[0206] It can be seen from Table 3 that, compared with the original simulation signal containing noise, the denoising method based on mode double decomposition proposed in the present invention can effectively improve the signal-to-noise ratio of the signal, which is closer to the pure signal; compared with the trend extraction method based on MWSA, wavelet denoising with improved threshold, hard threshold wavelet denoising and soft threshold wavelet denoising, the denoising signal obtained by the denoising method based on mode double decomposition has the highest signal-to-noise ratio and the lowest root mean square error, and the denoising effect is significantly improved.
[0207] The monitoring parameters of the exhaust temperature of the main steam turbine during the actual operation of a certain type of ship steam power system are taken for a period of 2000s, the sampling frequency is 1Hz, and the noise reduction method proposed by the present invention is adopted, and the effect is shown in Figure 13. Since it is actual data and a pure signal cannot be obtained, information entropy and power spectrum entropy are used as evaluation indicators, and the results are shown in Table 4:
[0208] Table 4 Comparison of denoising methods based on mode dual decomposition and other denoising methods
[0209]
[0210] It can be seen from Table 4 that, for the actual operation data, the denoising method based on modal double decomposition proposed in the present invention has the lowest information entropy and power spectrum entropy, which proves that the method also has certain superiority in practical applications and has practical engineering application value.
[0211] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for thermal monitoring parameter noise reduction based on modal double decomposition, It is characterized in that include: Perform MREMD based median mode decomposition on the operating parameters of the thermal system to obtain IMF components and residual components; Calculate the permutation entropy value of the IMF component based on the cross-correlation CC algorithm, perform K-means clustering on the permutation entropy value, and obtain a high-frequency IMF component and a low-frequency IMF component respectively; summing and reconstructing the high-frequency IMF components to obtain the noise components of the operating parameters of the thermal system; The VMD decomposition parameters are obtained by optimizing the NGO algorithm of the northern goshawk, and the VMD decomposition is performed on the noise component to obtain the IMF* component; The IMF* component with the lowest envelope entropy is used to reconstruct the residual component and the low-frequency IMF component to obtain the original data after noise reduction. The specific process includes: Calculating the envelope entropy of the IMF* component, and using the envelope entropy as the fitness of the NGO algorithm in the VMD decomposition parameter optimization process; When the envelope entropy of the IMF* component is the lowest, the fitness of the NGO algorithm is optimal, and the IMF* component at this time is used to reconstruct the residual component and the low-frequency IMF component.
2. According to the thermal monitoring parameter denoising method based on modal double decomposition according to claim 1, It is characterized in that The specific method for calculating the permutation entropy value of the IMF component based on the CC algorithm includes: The permutation entropy value is calculated for the IMF component, and the CC algorithm in chaotic system analysis is used to iteratively calculate and optimize the time delay and embedding dimension of the phase space reconstruction of the IMF component to obtain the permutation entropy value of the IMF component.
3. According to the thermal monitoring parameter denoising method based on modal double decomposition as described in claim 1, It is characterized in that The VMD decomposition parameters are obtained by optimizing the NGO algorithm by the Northern Goshawk, and the VMD decomposition is performed on the noise component to obtain the IMF* component. The specific method includes: The IMF components are decomposed and extracted using a variational mode decomposition (VMD) algorithm, and the obtained IMF components are recorded as IMF* components; The penalty factor and the number of decomposition layers of the VMD algorithm are optimized by using the NGO algorithm to obtain the optimal solution of the penalty factor and the number of decomposition layers of the VMD algorithm within a preset iteration range; The IMF* component is determined under the penalty factor and the number of decomposition levels of the optimal solution.
4. The thermal monitoring parameter denoising method based on modal dual decomposition according to claim 1, It is characterized in that Also includes: According to the sliding window method, the operating parameters of the thermal system are denoised within a preset window.
5. A thermal monitoring parameter noise reduction system based on modal double decomposition, It is characterized in that The method for reducing the noise of thermal monitoring parameters based on modal double decomposition as claimed in any one of claims 1 to 4 comprises: MREMD decomposition module, used to perform MREMD-based median mode decomposition on the operating parameters of the thermal system to obtain IMF components and residual components; An IMF component clustering module is used to calculate the permutation entropy value of the IMF component based on a cross-correlation CC algorithm, and perform K-means clustering on the permutation entropy value to obtain a high-frequency IMF component and a low-frequency IMF component respectively; A high-frequency noise reconstruction module, used for summing and reconstructing the high-frequency IMF components to obtain the noise components of the thermal system operation parameters; A VMD component decomposition module, used for obtaining VMD decomposition parameters by optimizing the NGO algorithm of the northern goshawk, and performing VMD decomposition on the noise component to obtain an IMF* component; The component reconstruction and denoising module is used to reconstruct the IMF* component with the lowest envelope entropy with the residual component and the low-frequency IMF component to obtain the original data after denoising, and is specifically used for: Calculating the envelope entropy of the IMF* component, and using the envelope entropy as the fitness of the NGO algorithm in the VMD decomposition parameter optimization process; When the envelope entropy of the IMF* component is the lowest, the fitness of the NGO algorithm is optimal, and the IMF* component at this time is used to reconstruct the residual component and the low-frequency IMF component.
6. The thermal monitoring parameter noise reduction system based on modal dual decomposition according to claim 5, It is characterized in that The IMF component clustering module is specifically used for: The permutation entropy value is calculated for the IMF component, and the CC algorithm in chaotic system analysis is used to iteratively calculate and optimize the time delay and embedding dimension of the phase space reconstruction of the IMF component to obtain the permutation entropy value of the IMF component.
7. The thermal monitoring parameter noise reduction system based on modal dual decomposition according to claim 5, It is characterized in that The VMD component decomposition module is specifically used for: The IMF components are decomposed and extracted using a variational mode decomposition (VMD) algorithm, and the obtained IMF components are recorded as IMF* components; The penalty factor and the number of decomposition layers of the VMD algorithm are optimized by using the NGO algorithm to obtain the optimal solution of the penalty factor and the number of decomposition layers of the VMD algorithm within a preset iteration range; The IMF* component is determined under the penalty factor and the number of decomposition levels of the optimal solution.
8. The thermal monitoring parameter noise reduction system based on modal dual decomposition according to claim 5, It is characterized in that It also includes a sliding window noise reduction module, which is used to reduce the noise of the thermal system operating parameters within a preset window according to the sliding window method.
