WO2024031938A1 - 一种基于isfo-vmd-kelm的sf6分解组分co2浓度反演的方法 - Google Patents

一种基于isfo-vmd-kelm的sf6分解组分co2浓度反演的方法 Download PDF

Info

Publication number
WO2024031938A1
WO2024031938A1 PCT/CN2023/073975 CN2023073975W WO2024031938A1 WO 2024031938 A1 WO2024031938 A1 WO 2024031938A1 CN 2023073975 W CN2023073975 W CN 2023073975W WO 2024031938 A1 WO2024031938 A1 WO 2024031938A1
Authority
WO
WIPO (PCT)
Prior art keywords
swordfish
concentration
optimizer
improved
kelm
Prior art date
Application number
PCT/CN2023/073975
Other languages
English (en)
French (fr)
Inventor
张英
黄杰
王为
王明伟
刘喆
冯楚杰
蒲曾鑫
赵世钦
潘云
Original Assignee
贵州电网有限责任公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 贵州电网有限责任公司 filed Critical 贵州电网有限责任公司
Publication of WO2024031938A1 publication Critical patent/WO2024031938A1/zh

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/39—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using tunable lasers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the present invention relates to the technical field of detection of internal decomposition components of SF 6 electrical equipment, and in particular to a method for inverting the concentration of CO 2 concentration of SF 6 decomposition components based on ISFO-VMD-KELM.
  • Solid insulating parts such as GIS basin insulators in SF 6 electrical equipment are cast from organic insulating materials such as epoxy resin.
  • Organic insulating materials will gradually crack and carbonize under the action of local high temperatures due to partial discharge or overheating faults, thus causing damage to the electrical equipment. The security of the entire device is threatened.
  • CO 2 gas is the decomposition product of organic insulating materials during the cracking and carbonization process. By detecting the volume fraction of CO 2 gas, insulation faults in SF 6 electrical equipment can be detected in time.
  • tunable laser absorption spectroscopy technology tunable absorption spectroscopy technology has been widely used in the power industry.
  • Tunable absorption spectroscopy technology utilizes the narrow spectral line width characteristics of tunable semiconductor lasers to achieve quantitative analysis of the target gas by observing the absorption spectral lines of the target gas to a specific laser. Therefore, the traditional SF 6 decomposition product concentration inversion method is The linear relationship between the volume fraction of the target gas and the absorbed light intensity is used to establish the least squares linear or nonlinear equation. Although this method is simple, it has low accuracy and large concentration inversion errors.
  • ELM Extreme Learning Machine
  • BP neural network to establish the concentration inversion model of O 2 gas
  • the genetic algorithm has weak local search ability and is prone to premature convergence.
  • KELM kernel extreme learning machine
  • the present invention provides an SF 6 decomposition component CO 2 concentration based on ISFO-VMD-KELM
  • the inversion method can solve problems such as the learning machine's strong dependence on parameters, its tendency to fall into local minima, its weak local search ability, its tendency to converge prematurely, and its particle swarm optimization algorithm's tendency to fall into local optimal solutions.
  • the present invention provides the following technical solution, a method for SF 6 decomposition component CO 2 concentration inversion based on ISFO-VMD-KELM, including:
  • the improved swordfish optimizer includes:
  • the fitness value will be calculated to continue optimization, and the final result will be returned when the iteration is over.
  • the improved swordfish optimizer also includes:
  • Tent chaos sequence is used to improve the initial population distribution of swordfish and sardines. Its mathematical expression is as follows:
  • ⁇ is a random number in [0,1].
  • the improved swordfish optimizer also includes:
  • u ⁇ N(0, ⁇ 2 ), v ⁇ N(0,1), and ⁇ are random numbers in [0,2].
  • the improved swordfish optimizer also includes: adaptive variational mode
  • the parameters of decomposition and kernel extreme learning machines were optimized, and Cauchy mutation and adaptive t-distribution mutation were performed on elite individuals of swordfish and sardine.
  • the inversion model includes:
  • optimization iterations are performed until the maximum number of iterations is reached, and the global optimal position in the last iteration is used as the optimal parameter for adaptive variational mode decomposition, and adaptive variational mode decomposition is performed under these parameter conditions.
  • the inversion model also includes using a kernel extreme learning machine to establish CO 2
  • the concentration prediction model is used, and the improved swordfish optimizer algorithm is used to optimize the regularization coefficient and kernel parameters of the kernel extreme learning machine.
  • the fitness function is the root mean square error of the predicted concentration and the true concentration.
  • the preprocessing includes using the improved swordfish optimizer to perform Adaptive variational mode decomposition after parameter optimization is used, combined with the wavelet threshold method to preprocess the original absorption spectrum.
  • the adaptive variational mode decomposition includes:
  • f(t) is a waveform function, that is, a time series, t is time, and a 1s waveform is collected
  • f 1 (t) is a 40Hz cosine signal with an amplitude of 1
  • f 2 (t) is a 40Hz cosine signal with an amplitude of 3 150Hz intermittent sinusoidal signal
  • f 3 (t) is 20dB Gaussian white noise.
  • the inversion model includes: dividing the second harmonic peak and the gas concentration respectively Serves as single input and single output of the model being built.
  • the present invention proposes a method for SF 6 decomposition component CO 2 concentration inversion based on ISFO-VMD-KELM.
  • the present invention adopts the adaptive variational mode decomposition joint wavelet optimized by the improved swordfish optimizer.
  • the threshold method filters the original spectrum signal to remove the high-frequency noise in the original signal, and then after SF 6 background subtraction, the amplitude of the spectrum can be read more accurately.
  • the CO 2 concentration inversion model is established using the kernel extreme learning machine optimized by the improved swordfish optimizer, which has higher accuracy and stability than the traditional concentration inversion method.
  • Figure 1 is a flow chart of a method for inverting the CO 2 concentration of the decomposed component of SF 6 based on ISFO-VMD-KELM provided by an embodiment of the present invention
  • Figure 2 is an adaptive variational mode decomposition mode diagram of an ISFO-VMD-KELM-based SF 6 decomposition component CO 2 concentration inversion method provided by an embodiment of the present invention
  • Figure 3 shows an SF 6 decomposition group based on ISFO-VMD-KELM provided by an embodiment of the present invention. Comparison chart of CO 2 absorption spectrum before and after filtering based on CO 2 concentration inversion method;
  • Figure 4 is a comparison chart of the SF 6 absorption spectrum before and after filtering based on the ISFO-VMD-KELM SF 6 decomposition component CO 2 concentration inversion method provided by one embodiment of the present invention
  • Figure 5 is a background-subtracted absorption spectrum diagram of a method for inverting the CO 2 concentration of the SF 6 decomposition component based on ISFO-VMD-KELM provided by one embodiment of the present invention
  • Figure 6 is a background-subtracted absorption spectrum diagram of a method for inverting the CO 2 concentration of the decomposed component of SF 6 based on ISFO-VMD-KELM provided by one embodiment of the present invention
  • Figure 7 is a least squares linear fitting of a method for inverting the CO 2 concentration of the SF 6 decomposition component based on ISFO-VMD-KELM provided by one embodiment of the present invention
  • references herein to "one embodiment” or “an embodiment” refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. "In one embodiment” appearing in different places in this specification does not all refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.
  • connection should be understood in a broad sense.
  • it can be a fixed connection, a detachable connection, or an integrated connection; it can also be a mechanical connection, an electrical connection, or a direct connection.
  • a connection can also be indirectly connected through an intermediary, or it can be an internal connection between two components.
  • the specific meanings of the above terms in the present invention can be understood on a case-by-case basis.
  • This embodiment provides a method for inverting the concentration of CO 2 concentration of SF 6 decomposition components based on ISFO-VMD-KELM, including:
  • S1 Based on the traditional Sailfish optimizer, improve the optimization accuracy and local search capabilities, and establish an improved Sailfish optimizer;
  • the Sailfish Optimizer is inspired by the natural phenomenon of swordfish preying on sardines in marine ecology: swordfish adopt a group cooperative hunting method, forcing sardine groups to surface by driving them to the surface before starting to hunt. After the swordfish attacks, Injured sardines constantly break away from the team and are preyed upon by swordfish.
  • i represents the current iteration number
  • Represents the updated position of the swordfish Indicates the position of the leader swordfish.
  • the leader swordfish is the swordfish with the best fitness value currently.
  • the injured sardine is the sardine that currently has the optimal fitness value.
  • r represents a random number between (0,1)
  • ⁇ i represents the iteration coefficient.
  • PD represents the density of sardine groups, and the mathematical expression of PD is as follows:
  • N SF and N S represent the population numbers of swordfish and sardine in each iteration respectively.
  • A represents the maximum attack strength of the swordfish
  • ⁇ represents the attenuation coefficient
  • di represents the dimension in the i-th iteration
  • the number of parameters that need to be optimized is the number of dimensions
  • d represents the dimension, that is, the d-th parameter
  • i represents the number of iterations
  • di represents the d-dimensional parameter in the i-th iteration. (variable).
  • f(SF i ) represents the current fitness value of swordfish
  • f(S i ) represents the current fitness value of sardine
  • the swordfish optimizer uses a pseudo-random method to generate the initial population. This method cannot guarantee the balance of the population distribution, and can easily cause the search state to stagnate before the optimal solution is found.
  • Chaos has the characteristics of determinism and quasi-randomness.
  • Figure 2 shows the probability distribution histograms of four common chaotic sequences. Among them, the Tent sequence has the most uniform distribution on [0,1]. Therefore, the present invention uses the Tent chaotic sequence to To improve the initial population distribution of swordfish and sardines, the mathematical expression is as follows:
  • ⁇ is a random number in [0,1], and the present invention takes the value 0.7.
  • the lens imaging learning mechanism is further combined to screen high-quality populations.
  • Both swordfish and sardines generate initial populations according to the following steps:
  • Tent chaos mapping is used to generate N population individuals, recorded as population A, and their fitness values are calculated;
  • xi represents the current solution
  • N populations (number value) a and b represent the minimum and maximum values of the population
  • k is 1, a i and b i represent the minimum and maximum boundaries of the population respectively.
  • the sardine update method does not have enough search power in the late iteration period, so it is easy to fall into the local optimal solution.
  • the Levy flight strategy is based on the change of the random step size to make a position mutation movement corresponding to the distance. Therefore, in order to increase the search range of the sardine and improve the optimization performance, the present invention introduces the Levy random step formula into the position update method of the sardine. Increase the search range of sardines, and the improved mathematical expression of sardine position update is as follows.
  • the sardine update method of the standard swordfish optimization algorithm does not have sufficient search strength in the late iteration period, so it is easy to fall into the local optimal solution. Therefore, the present invention improves the sardine position update method. , while introducing an adaptive feedback factor to accelerate the iterative convergence process.
  • u ⁇ N(0, ⁇ 2 ), v ⁇ N(0,1), and ⁇ are random numbers in [0,2], which are taken as 1.5 in this patent.
  • the optimal solution is used as the elite, and the mutation operation is performed as follows:
  • cauchy(0,1) is a 0-1 random number obeying the Cauchy distribution, Expresses elite solution Solution after Cauchy mutation.
  • t(i) represents the Student t distribution, and its degree of freedom is the number of iterations of the algorithm.
  • the elite solution is defined here as the optimal solution (calculated through the fitness function value, which will have the optimal fitness value position vectors are called elites).
  • the fitness value will be calculated to continue optimization, and the final result will be returned when the iteration is over.
  • S2 Collect original absorption spectra, and preprocess the original absorption spectra in combination with the variational mode decomposition algorithm optimized by the improved swordfish optimizer and the wavelet threshold method;
  • variational mode decomposition is a completely non-recursive modal variation method. It decomposes the original signal into several eigenmodes that meet the constraint conditions through iterative optimization, effectively avoiding the problem of processing non-linear modes similar to EMD. Modal aliasing and end-effect problems that occur when smoothing vibration signals.
  • variational mode decomposition requires the setting of four control parameters, which are highly dependent on parameters. Among them, K determines the number of decomposition layers, and the penalty factor ⁇ determines the spectrum bandwidth of the decomposition mode. Different (K, ⁇ ) combinations have The final decomposition effect will have varying degrees of impact. To this end, the present invention proposes to use an improved swordfish optimizer to perform two-dimensional optimization of (K, ⁇ ) of variational mode decomposition to improve the adaptive ability and decomposition effect of variational mode decomposition.
  • the swordfish optimizer is only used to optimize parameters, and the kernel extreme learning machine (KELM) is the algorithm for establishing the CO2 concentration model.
  • KELM kernel extreme learning machine
  • the collected spectral samples are divided into training sets and test sets, and the training set is used as a sample for training the model.
  • the peak of the spectrum is used as a single input of KELM, and the concentration is used as a single output. After training, a fixed model is obtained, and a corresponding output concentration (predicted value) is obtained by inputting the harmonic peak.
  • the root mean square error between the predicted value and the actual value is used as the fitness function, and the regularization coefficient and kernel function parameters of KELM are iteratively optimized through the ISSFO algorithm to obtain the parameters that minimize the prediction error of KELM, and then substituted into the parameters to establish CO2 concentration model.
  • the test set is used to test the effect of the model.
  • the improved Sailfish optimizer can search for the optimal fitness value faster (fewer iterations) during the optimization process, and the optimization accuracy is higher (the fitness value is smaller).
  • the present invention uses multi-scale permutation entropy (MPE) as the fitness function when the improved swordfish optimizer algorithm optimizes the two-dimensional parameters of variational mode decomposition.
  • Permutation entropy PE
  • PE Permutation entropy
  • An evaluation index of degree The larger the entropy value, the higher the degree of data chaos. On the contrary, it means the lower the randomness of the data.
  • Multi-scale permutation entropy (MPE) first coarse-grains the time series data and then calculates the permutation entropy.
  • Variational mode decomposition is performed under the condition of (K, ⁇ ) parameters.
  • test signal as shown in the following formula was constructed in MATLAB, a sampling rate of 1kHz was set, and 1000 sample points were collected in 1 second through simulation.
  • f(t) is a waveform function, that is, a time series, t is time, and a 1s waveform is collected
  • f 1 (t) is a 40Hz cosine signal with an amplitude of 1
  • f 2 (t) is a 40Hz cosine signal with an amplitude of 3 150Hz intermittent sinusoidal signal
  • f 3 (t) is 20dB Gaussian white noise.
  • the population number of the improved swordfish optimizer algorithm is set to 30, of which the sardine population accounts for 70%, the swordfish population accounts for 30%, the search dimension is 2 dimensions, and the search boundary of the variational mode decomposition decomposition layer K is [3,9], the search boundary of the penalty factor ⁇ is [50,5000], and both K and ⁇ are rounded.
  • the optimal parameter was obtained as (3,228).
  • the signal was decomposed into three eigenmodes.
  • Figure 2 shows the time domain sum corresponding to each decomposed mode.
  • Spectrogram in which the center frequencies of IMF1 and IMF2 are 40Hz and 150Hz respectively.
  • the two modes correspond to cosine signals and intermittent sine signals respectively.
  • IMF3 is superimposed Gaussian white noise.
  • the present invention uses tunable absorption spectroscopy technology combined with wavelength modulation spectroscopy (WMS) as the experimental detection method.
  • WMS wavelength modulation spectroscopy
  • I 0 represents the initial light intensity of the laser with wave number v
  • I represents the light intensity after passing through the gas to be measured
  • ⁇ (v) represents the absorption coefficient of the gas to be measured
  • P represents the gas pressure
  • C represents the gas volume fraction
  • L represents Gas absorption chamber length.
  • the amplitude I 2f of the second harmonic signal is proportional to the volume fraction C of the measured gas, that is:
  • ⁇ 0 is the absorption cross section, and the absorption coefficient ⁇ and ⁇ 0 can be converted into each other.
  • the light source uses a semiconductor laser (VERTILAS, VL-2004-1m) with an emission wavelength of 2004nm, and the maximum output power is 3mW.
  • the laser wavelength is controlled by adjusting the laser temperature and driving current.
  • This experimental device uses the TEC driver chip MAX1968 to control the laser temperature.
  • the driving current is a superimposed signal of a 40Hz triangle wave and a 72kHz sine wave.
  • the modulated laser is collimated by a lens (THORLABS, C220TMD) and then passes through the absorption gas chamber (length is 25cm).
  • the multi-strategy method of Tent chaos mapping combined with lens imaging learning provides a more uniform and diverse population for the swordfish optimizer algorithm, and the sardines integrated with levy random steps have better Optimization performance, adaptive t-distribution mutation and Cauchy mutation effectively improve the ability of the swordfish optimizer algorithm to jump out of the local optimum.
  • the improved swordfish optimizer algorithm is used to optimize the decomposition layer number K and penalty factor ⁇ of variational mode decomposition, which overcomes the need for variational mode decomposition parameters.
  • the artificially set problem improves the adaptive ability of variational mode decomposition.
  • the filtering method of variational mode decomposition combined with wavelet threshold can effectively filter the interference noise in the absorption spectrum, read the second harmonic amplitude more accurately, and improve detection accuracy.
  • an embodiment of the present invention provides a method for inverting the CO 2 concentration of the decomposed component of SF 6 based on ISFO-VMD-KELM.
  • a comparative experiment is conducted scientific argument.
  • this experiment uses SF 6 as the background gas and CO 2 as the target detection gas component.
  • the SF 6 dynamic gas distribution system is used to configure SF 6 and CO 2 Mixed sample gas, configured CO2 sample gas volume fraction includes 0.87%, 0.85%, 0.83%, 0.8%, 0.77%, 0.75%, 0.72%, 0.7%, 0.68%, 0.65%, 0.62%, 0.6%, 0.57%, 0.55%, 0.53%, 0.5%.
  • the above configured gases were filled into the tunable absorption spectrum technology detection device one by one for experiments. 60 second harmonic spectrum data were collected for each concentration group, and a total of 960 spectra were collected.
  • the adaptive variational mode decomposition proposed above is used to decompose the original spectrum, and the mode with high correlation is selected for reconstruction according to the correlation coefficient between each decomposed mode and the original data.
  • the wavelet threshold method is used to reconstruct the reconstructed signal. Filter again to obtain the absorption spectral lines after removing noise.
  • Figures 4 and 5 are respectively comparisons of the absorption spectra of pure SF 6 and CO 2 with a volume fraction of 0.85% before and after filtering.
  • the processed spectral signal not only removes the high-frequency noise in the original signal, but also has a smoother waveform.
  • the wavelength modulation spectroscopy method is different from the direct absorption spectrometry method that needs to fit the background baseline.
  • the background signal is directly subtracted from the absorption spectrum line of the target gas.
  • subtracting the SF 6 absorption spectrum line in Figure 5 from the CO 2 absorption spectrum line in Figure 4 can obtain the second harmonic absorption spectrum line of CO 2 with a volume fraction of 0.85% after subtracting the background.
  • the interference harmonic components around the center frequency of the second harmonic after subtracting the background are effectively weakened, making it easier to accurately read the amplitude of the second harmonic.
  • Figure 7 shows 16 preprocessed sets of CO 2 absorption spectra of different concentrations.
  • Table 1 shows the RMSE comparison of the concentration inversion results:
  • the second harmonic amplitude is proportional to the measured gas concentration. Therefore, the most commonly used concentration inversion method in the power industry is to use the least squares method to establish a linear relationship between the second harmonic amplitude and the gas concentration.
  • the present invention chooses to use the kernel extreme learning machine (KELM) to establish the CO 2 concentration prediction model, and uses the improved swordfish optimizer algorithm to optimize the regularization coefficient C and kernel parameter S of the kernel extreme learning machine.
  • the fitness function is predicted Root mean square error (RMSE) between concentration and true concentration.
  • the improved SFO-KELM and SFO-KELM, ELM Sigmoidal activation function, hidden layer nodes are 20
  • GA-BP the crossover probability of GA is 0.7 , the mutation probability is 0.3; the number of BP training times is 100, the training efficiency is 0.01, and the hidden layer nodes are 20
  • PSO-KELM the inertia factor w of PSO is 1, and the acceleration factors C1 and C2 are both 2)
  • CLS Model accuracy comparison The experimental results of 30 iterations of each algorithm are shown in Table 1.
  • the root mean square error (RMSE) of ISFO-KELM in Table 1 is the smallest.
  • the root mean square error of the training set of ISFO-KELM is 30.8% smaller than PSO-KELM and 85.4% smaller than SFO-KELM.
  • the test set root mean square error of ISFO-KELM is 36.9% smaller than PSO-KELM and 86.8% smaller than SFO-KELM.
  • the root mean square error of the training set and test set of ISFO-KELM is 2 orders of magnitude smaller than GA-BP and ELM, and 3 orders of magnitude smaller than the linear least squares (CLS) method. Therefore, the kernel extreme learning machine optimized by the improved Sailfish optimizer algorithm has excellent performance in accuracy.
  • embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment that combines software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • the solutions in the embodiments of this application can be implemented using various computer languages, such as the object-oriented programming language Java and the literal scripting language JavaScript.
  • These computer program instructions may also be stored in a computer-readable memory that causes a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction means, the instructions
  • the device implements the functions specified in a process or processes of the flowchart and/or a block or blocks of the block diagram.
  • These computer program instructions may also be loaded onto a computer or other programmable data processing device, causing a series of operating steps to be performed on the computer or other programmable device to produce computer-implemented processing, thereby executing on the computer or other programmable device.
  • Instructions provide steps for implementing the functions specified in a process or processes of a flowchart diagram and/or a block or blocks of a block diagram.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Chemical & Material Sciences (AREA)
  • Optics & Photonics (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

一种基于ISFO-VMD-KELM的SF 6分解组分CO 2浓度反演的方法，包括：基于传统旗鱼优化器，建立改进旗鱼优化器（S1）；收集原始吸收光谱，对原始吸收光谱进行预处理（S2）；根据预处理结果，结合改进旗鱼优化器优化的核极限学习机建立吸收光谱的二次谐波幅值与CO 2浓度的反演模型（S3）；根据反演模型进行CO 2检测实验，判断SF 6电气设备的运行状态（S4）。采用改进旗鱼优化器优化的自适应变分模态分解联合小波阈值法对原始光谱信号进行滤波，去除了原始信号中的高频噪声，再经过SF 6背景扣除后能够更准确地读取光谱的幅值。采用改进旗鱼优化器优化的核极限学习机建立CO 2浓度反演模型，相比传统的浓度反演方法具有更高的精度和稳定性。

Description

一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法 技术领域
本发明涉及SF6电气设备内部分解组分检测技术领域，尤其涉及一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法。
背景技术
SF6电气设备中如GIS盆式绝缘子等固体绝缘件是由环氧树脂等有机绝缘材料浇注而成，有机绝缘材料在局部放电或过热性故障的局部高温作用下会逐渐裂解和碳化，从而对整个设备的安全构成威胁。CO2气体是有机绝缘材料在裂解和碳化过程中的分解产物，通过检测CO2气体的体积分数能及时发现SF6电气设备的绝缘性故障。近年来，可调谐激光吸收光谱技术(可调谐吸收光谱技术)在电力行业被广泛应用。可调谐吸收光谱技术技术是利用可调谐半导体激光器谱线宽度窄的特性，通过观察目标气体对特定激光的吸收谱线实现对目标气体的定量分析，因此传统的SF6分解产物浓度反演方法是利用目标气体的体积分数与吸收光强的线性关系建立最小二乘法线性或非线性方程式，该方法虽然简便，但是精度不高，浓度反演误差较大。
随着机器学习的发展，越来越多的研究将机器学习应用到可调谐吸收光谱技术技术中。现有技术采用极限学习机(Extreme Learning Machine,ELM)建立谐波信号与积分吸光度的模型，但ELM对参数依赖性强，容易陷入局部最小值。还有的现有技术采用BP神经网络建立O2气体的浓度反演模型，并在BP网络的参数调节问题上应用了经典遗传算法，但BP网络本身存在学习过程缓慢以及稳定性差的问题，而且遗传算法局部搜索能力较弱，容易过早收敛。还有些采用粒子群优化算法优化核极限学习机(KELM)，建立了精度较高的CO2浓度反演模型，但是粒子群优化算法容易陷入局部最优解。
发明内容
本部分的目的在于概述本发明的实施例的一些方面以及简要介绍一些较佳实施例。在本部分以及本申请的说明书摘要和发明名称中可能会做些简化或省略以避免使本部分、说明书摘要和发明名称的目的模糊，而这种简化或省略不能用于限制本发明的范围。
鉴于上述现有存在的问题，提出了本发明。
因此，本发明提供了一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度 反演的方法，能够解决学习机对参数依赖性强，容易陷入局部最小值、局部搜索能力较弱，容易过早收敛以及粒子群优化算法容易陷入局部最优解等问题。
为解决上述技术问题，本发明提供如下技术方案，一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，包括：
基于传统旗鱼优化器，改进寻优精度和局部搜索能力，建立改进旗鱼优化器；
收集原始吸收光谱，并结合所述改进旗鱼优化器优化的变分模态分解算法联合小波阈值法，对所述原始吸收光谱进行预处理；
根据所述预处理结果，结合所述改进旗鱼优化器优化的核极限学习机建立吸收光谱的二次谐波幅值与CO2浓度的反演模型；
根据所述反演模型进行CO2检测实验，判断SF6电气设备的运行状态。
作为本发明所述的基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的一种优选方案，其中：所述改进旗鱼优化器包括，
设置算法控制参数，根据多策略初始化方式生成初始旗鱼和沙丁鱼种群；
计算适应度值，记录本次迭代时的旗鱼和沙丁鱼全局最优位置；
更新旗鱼位置，根据攻击力度更新沙丁鱼位置；
分别对旗鱼和沙丁鱼的精英个体做柯西变异和自适应t分布变异；
比较旗鱼和沙丁鱼最优解，根据结果替换旗鱼和沙丁鱼位置；
迭代未结束则计算适应度值继续优化，迭代结束则返回最终结果。
作为本发明所述的基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的一种优选方案，其中：所述改进旗鱼优化器还包括，
采用Tent混沌序列来改善旗鱼和沙丁鱼的初始种群分布，其数学表达式如下：
其中，β是[0,1]的随机数。
作为本发明所述的基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的一种优选方案，其中：所述改进旗鱼优化器还包括，
改进后的沙丁鱼位置更新数学表达式如下：
Levy＝u/|v|1/β
其中σ的数学表达式为：
其中，u～N(0,σ2)，v～N(0,1)，β是[0,2]的随机数。
作为本发明所述的基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的一种优选方案，其中：所述改进旗鱼优化器还包括，对自适应变分模态分解和核极限学习机的参数进行寻优，对旗鱼和沙丁鱼的精英个体进行柯西变异和自适应t分布变异。
作为本发明所述的基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的一种优选方案，其中：所述反演模型包括，
初始化改进旗鱼优化器基本参数，将自适应变分模态分解的多尺度排列熵之和作为适应度函数；
生成初始种群，设置旗鱼种群和沙丁鱼种群比例；
将各种群位置信息作为参数导入适应度函数计算多尺度排列熵，并将最小多尺度排列熵值和对应位置向量作为全局解和全局最优位置；
更新旗鱼和沙丁鱼位置，重新计算适应度值，更新全局解和全局最优位置；
进行优化迭代，直到达到最大迭代次数，将最后一次迭代时的全局最优位置作为自适应变分模态分解最佳参数，在此参数条件下进行自适应变分模态分解。
作为本发明所述的基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的一种优选方案，其中：所述反演模型还包括，采用核极限学习机来建立CO2的浓度预测模型，并利用改进旗鱼优化器算法优化核极限学习机的正则化系数及核参数，适应度函数为预测浓度和真实浓度的均方根误差。
作为本发明所述的基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的一种优选方案，其中：所述预处理包括，利用所述改进旗鱼优化器进行 参数优化后的自适应变分模态分解，结合小波阈值法对原始吸收光谱进行预处理。
作为本发明所述的基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的一种优选方案，其中：所述自适应变分模态分解包括，
其中，f(t)是波形函数，即时间序列，t是时间，采集了1s的波形，f1(t)是幅值为1的40Hz余弦信号，f2(t)是幅值为3的150Hz间歇正弦信号，f3(t)是20dB的高斯白噪声。
作为本发明所述的基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的一种优选方案，其中：所述反演模型包括，将二次谐波峰值和气体浓度分别作为所建模型的单输入和单输出。
本发明的有益效果：本发明提出一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，本发明采用改进旗鱼优化器优化的自适应变分模态分解联合小波阈值法对原始光谱信号进行滤波，去除了原始信号中的高频噪声，再经过SF6背景扣除后能够更准确地读取光谱的幅值。采用改进旗鱼优化器优化的核极限学习机建立CO2浓度反演模型，相比传统的浓度反演方法具有更高的精度和稳定性。
附图说明
为了更清楚地说明本发明实施例的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其它的附图。其中：
图1为本发明一个实施例提供的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法流程图；
图2为本发明一个实施例提供的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的自适应变分模态分解模态图；
图3为本发明一个实施例提供的一种基于ISFO-VMD-KELM的SF6分解组 分CO2浓度反演的方法的CO2吸收光谱滤波前后对比图；
图4为本发明一个实施例提供的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的SF6吸收光谱滤波前后对比图；
图5为本发明一个实施例提供的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的扣除背景后的吸收谱线图；
图6为本发明一个实施例提供的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的扣除背景后的吸收谱线图；
图7为本发明一个实施例提供的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法的最小二乘法线性拟合；
具体实施方式
为使本发明的上述目的、特征和优点能够更加明显易懂，下面结合说明书附图对本发明的具体实施方式做详细的说明，显然所描述的实施例是本发明的一部分实施例，而不是全部实施例。基于本发明中的实施例，本领域普通人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本发明的保护的范围。
在下面的描述中阐述了很多具体细节以便于充分理解本发明，但是本发明还可以采用其他不同于在此描述的其它方式来实施，本领域技术人员可以在不违背本发明内涵的情况下做类似推广，因此本发明不受下面公开的具体实施例的限制。
其次，此处所称的“一个实施例”或“实施例”是指可包含于本发明至少一个实现方式中的特定特征、结构或特性。在本说明书中不同地方出现的“在一个实施例中”并非均指同一个实施例，也不是单独的或选择性的与其他实施例互相排斥的实施例。
本发明结合示意图进行详细描述，在详述本发明实施例时，为便于说明，表示器件结构的剖面图会不依一般比例作局部放大，而且所述示意图只是示例，其在此不应限制本发明保护的范围。此外，在实际制作中应包含长度、宽度及深度的三维空间尺寸。
同时在本发明的描述中，需要说明的是，术语中的“上、下、内和外”等指示的方位或位置关系为基于附图所示的方位或位置关系，仅是为了便于描述本发明和简化描述，而不是指示或暗示所指的装置或元件必须具有特定的方位、 以特定的方位构造和操作，因此不能理解为对本发明的限制。此外，术语“第一、第二或第三”仅用于描述目的，而不能理解为指示或暗示相对重要性。
本发明中除非另有明确的规定和限定，术语“安装、相连、连接”应做广义理解，例如：可以是固定连接、可拆卸连接或一体式连接；同样可以是机械连接、电连接或直接连接，也可以通过中间媒介间接相连，也可以是两个元件内部的连通。对于本领域的普通技术人员而言，可以具体情况理解上述术语在本发明中的具体含义。
实施例1
参照图1-2，为本发明的第一个实施例，该实施例提供了一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，包括：
S1：基于传统旗鱼优化器，改进寻优精度和局部搜索能力，建立改进旗鱼优化器；
更进一步的，旗鱼优化器(SFO)受海洋生态中旗鱼捕食沙丁鱼的自然现象启发：旗鱼采取群体协作的狩猎方式，通过驱赶迫使沙丁鱼群上浮至水面后才展开围猎，在旗鱼攻击下不断有受伤的沙丁鱼脱离队伍从而被旗鱼捕食。
应说明的是，旗鱼作为捕食者，其位置更新数学表达式如下：
其中，i表示当前迭代次数，表示旗鱼更新后的位置，表示领头旗鱼的位置，领头旗鱼是目前拥有最优适应度值的旗鱼，表示受伤沙丁鱼的位置，受伤沙丁鱼是目前拥有最优适应度值的沙丁鱼，表示当前旗鱼的位置，r表示(0,1)之间的随机数，λi表示迭代系数。
应说明的是，迭代系数与沙丁鱼种群密度相关，λi数学表达式如下：
λi＝2×r×PD-PD
其中，PD表示沙丁鱼群的密度，PD数学表达式如下：
其中，NSF、NS分别表示每次迭代中旗鱼和沙丁鱼种群数量。
更进一步的，沙丁鱼的移动方式与旗鱼的攻击力度相关，其位置更新数学表达式如下：
其中，表示沙丁鱼更新后的位置，表示当前沙丁鱼的位置，AP表示 每次迭代时旗鱼的攻击力度，AP的数学表达式如下：
AP＝A×(1-(2×i×ε))
其中，A表示旗鱼的最大攻击力度，ε表示衰减系数，在迭代过程中旗鱼的攻击力度从最大值A线性衰减到0。
应说明的是，当AP≥0.5时所有沙丁鱼需更新当前位置前往安全区域，当AP＜0.5时仅部分沙丁鱼选择更新当前位置。部分更新时需要考虑数量α和维度β，更新范围定义如下：
其中，di表示第i次迭代时的维度，需要优化的参数个数即维度数，d表示维度，即第d个参数，i表示迭代次数，di表示第i次迭代时的第d维参数(变量)。。
更进一步的，每次迭代时比较两个种群的适应度值，并根据公式更替旗鱼和沙丁鱼的位置。
其中，f(SFi)表示当前旗鱼的适应度值，f(Si)表示当前沙丁鱼的适应度值。
更进一步的，旗鱼优化器采用伪随机方式生成初始种群，该方法不能保证种群分布的均衡性，容易导致在未找到最优解时搜索状态就已陷入停滞。混沌具有确定性和类随机性的特点，图2展示了四种常见的混沌序列的概率分布直方图，其中Tent序列在[0,1]上的分布最均匀，因此本发明采用Tent混沌序列来改善旗鱼和沙丁鱼的初始种群分布，其数学表达式如下：
其中，β是[0,1]的随机数，本发明取值0.7.
更进一步的，在得到初始种群的基础上进一步结合透镜成像学习机制筛选优质种群。旗鱼和沙丁鱼均按如下步骤生成初始种群：
更进一步的，利用Tent混沌映射生成N个种群个体，记为种群A，并计算其适应度值；
更进一步的，利用上述公式，求取种群A中所有个体的透镜成像解，记为种群B，并计算其适应度值：
其中，xi表示当前解，表示当前解的镜像解，根据上式得到N个种群(数 值)，a和b表示种群最小值和最大值，根据经验值k取1，ai和bi分别表示种群的最小边界和最大边界。
应说明的是，比较适应度值，若透镜成像解优于其原解，则将种群A中的原解替换成B中对应的透镜成像解，最终组成N个新的初始种群。
更进一步的，沙丁鱼更新方式在迭代后期搜索力度不足，从而容易陷入局部最优解。
应说明的是，Levy飞行策略是根据随机步长的变化做相应距离的位置突变运动，因此为增加沙丁鱼的搜索范围提高寻优性能，本发明将levy随机步长式引入沙丁鱼的位置更新方式，增加沙丁鱼的搜索范围，改进后的沙丁鱼位置更新数学表达式如下标准旗鱼优化算法的沙丁鱼更新方式在迭代后期搜索力度不足，从而容易陷入局部最优解，因此本发明对沙丁鱼位置更新方式进行改进，同时引入一种自适应反馈因子加速迭代收敛过程。
更进一步的，当AP≥0.5时，改进后的沙丁鱼位置更新数学表达式如下：
Levy＝u/|v|1/β
其中σ的数学表达式为：
其中，u～N(0,σ2)，v～N(0,1)，β是[0,2]的随机数，本专利中取1.5.
更进一步的，将最优解作为精英，按如下步骤进行变异操作：
分别对旗鱼和沙丁鱼的精英个体进行柯西变异，数学表达式如下：
其中，cauchy(0,1)为服从柯西分布的0-1随机数，表示精英解柯西变异后的解。
分别对旗鱼和沙丁鱼精英个体进行自适应t分布变异，数学表达式如下：
其中，表示精英解的t分布变异解，t(i)表示学生t分布，其自由度是算法的迭代次数。
应说明的是，比较精英解与其变异解的适应度值，选择较优的变异解替换对应精英解。学生t-分布(Student's t-distribution)，可简称为t分布。精英解这里定义为最优解(通过适应度函数值计算，将拥有最优适应度值得位置 向量称为精英)。
更进一步的，改进旗鱼优化器的步骤如下：
设置算法控制参数，根据多策略初始化方式生成初始旗鱼和沙丁鱼种群；
计算适应度值，记录本次迭代时的旗鱼和沙丁鱼全局最优位置；
更新旗鱼位置，根据攻击力度更新沙丁鱼位置；
分别对旗鱼和沙丁鱼的精英个体做柯西变异和自适应t分布变异；
比较旗鱼和沙丁鱼最优解，根据结果替换旗鱼和沙丁鱼位置；
迭代未结束则计算适应度值继续优化，迭代结束则返回最终结果。
S2：收集原始吸收光谱，并结合所述改进旗鱼优化器优化的变分模态分解算法联合小波阈值法，对所述原始吸收光谱进行预处理；
更进一步的，变分模态分解是一种完全非递归的模态变分方法，通过迭代寻优将原始信号分解为若干个满足约束条件的本征模态，有效避免了类似EMD在处理非平稳振动信号时出现的模态混叠和端点效应问题。
应说明的是，变分模态分解需要设置四个控制参数，对参数依赖程度高，其中K决定分解层数，惩罚因子α决定分解模态的频谱带宽，不同的(K,α)组合对最后呈现的分解效果会有不同程度的影响。为此，本发明提出利用改进的旗鱼优化器对变分模态分解的(K,α)进行二维寻优，提高变分模态分解的自适应能力和分解效果。
旗鱼优化器只是优化参数用的，核极限学习机(KELM)才是建立CO2浓度模型的算法。采集的光谱样本划分成训练集和测试集，将训练集作为训练模型的样本。光谱的峰值作为KELM的单输入，浓度作为单输出，经过训练后得到一个固定的模型，输入谐波峰值便得到一个对应的输出浓度(预测值)。将预测值与实际值的均方根误差作为适应度函数，通过ISSFO算法对KELM的正则化系数和核函数参数进行迭代寻优，得到使KELM的预测误差最小的参数，然后再代入参数建立CO2浓度模型。测试集是用来检验模型的效果的。
改进旗鱼优化器相对于传统的旗鱼优化器，在寻优过程中能更快搜索到最优适应度值(迭代次数更少)，并且寻优精度更高(适应度值更小)。
更进一步的，本发明采用多尺度排列熵(MPE)作为改进旗鱼优化器算法优化变分模态分解二维参数时的适应度函数。排列熵(PE)常被用做判断时序数据混乱 程度的评价指标，熵值越大表示数据混乱程度越高，反之则表示数据随机性越低，而多尺度排列熵(MPE)则是先将时序数据粗粒化后再计算排列熵。
更进一步的，完整的优化步骤如下：
初始化改进旗鱼优化器基本参数，将变分模态分解各分解模态的排列熵之和作为适应度函数，设置K的搜索边界为[3,9]，α的搜索边界为[50,5000]；
生成初始种群，设置旗鱼种群和沙丁鱼种群比例为3比7；
将各种群位置信息作为参数(K,α)导入适应度函数计算排列熵，并将最小排列熵值和对应位置向量作为全局解和全局最优位置；
根据上文定义更新旗鱼和沙丁鱼位置；
重新计算适应度值，更新全局解和全局最优位置；
进行优化迭代，直到达到最大迭代次数；
将最后一次迭代时的全局最优位置(K,α)作为变分模态分解最佳参数；
在(K,α)参数条件下进行变分模态分解分解。
应说明的是，为测试改进变分模态分解的分解效果，在MATLAB中构造了如下式所示的测试信号，设置1kHz的采样率，仿真1秒采集到1000个样本点。
其中，f(t)是波形函数，即时间序列，t是时间，采集了1s的波形，f1(t)是幅值为1的40Hz余弦信号，f2(t)是幅值为3的150Hz间歇正弦信号，f3(t)是20dB的高斯白噪声。
更进一步的，设置改进旗鱼优化器算法种群数为30，其中沙丁鱼种群占比70％，旗鱼种群占比30％，搜索维度为2维，变分模态分解分解层数K的搜索边界为[3,9]，惩罚因子α的搜索边界为[50,5000]，K和α均做取整处理。
应说明的是，经过数次迭代寻优后，得到最佳参数为(3,228)，在该参数条件下信号被分解为3个本征模态，图2为各分解模态对应的时域和频谱图，其中IMF1和IMF2的中心频率分别为40Hz、150Hz，两个模态分别对应余弦信号和间歇式正弦信号，IMF3则是叠加的高斯白噪声。
S3：根据所述预处理结果，结合所述改进旗鱼优化器优化的核极限学习机建立吸收光谱的二次谐波幅值与CO2浓度的反演模型；
更进一步的，本发明采用可调谐吸收光谱技术联合波长调制光谱法(WMS)作为实验检测方法，根据气体的红外光谱吸收原理，气体分子对特定波长的光具有选择吸收特性，由Beer-Lambert定律解释为：
I＝I0exp(-α(v)PCL)
其中，I0表示波数为v的激光器初始光强，I表示穿过被测气体后的光强，α(v)表示被测气体吸收系数，P表示气体压强，C表示气体体积分数，L表示气体吸收气室长度。
更进一步的，根据傅里叶变换原理将上式展开后发现二次谐波信号的幅值I2f与被测气体的体积分数C成正比关系，即：
I2f∝I0σ0CL
应说明的是，δ0是吸收截面，吸收系数α和δ0可相互转换。光源采用发射波长为2004nm的半导体激光器(VERTILAS，VL-2004-1m)，最大输出功率为3mW。激光器波长是通过调节激光器温度和驱动电流控制的，本实验装置采用TEC驱动芯片MAX1968控制激光器温度，驱动电流为40Hz三角波与72kHz正弦波的叠加信号。调制激光经透镜(THORLABS，C220TMD)准直后穿过吸收气室(长度为25cm)，被气室内被测气体选择吸收后由气室另一侧的光电探测器(滨松，12181-020)接收，然后通过锁相放大器提取出吸收光谱的二次谐波信号。可调谐吸收光谱技术实验装置采集的原始二次谐波信号在PC端的可调谐吸收光谱技术检测软件中进行数据处理。
S4：根据所述反演模型进行CO2检测实验，判断SF6电气设备的运行状态。
更进一步的，旗鱼优化器算法的改进策略中，Tent混沌映射联合透镜成像学习的多策略方法为旗鱼优化器算法提供了更均匀多样的种群，融合levy随机步长的沙丁鱼具有更好的寻优性能，自适应t分布变异和柯西变异有效提高了旗鱼优化器算法跳出局部最优的能力。
应说明的是，根据多尺度排列熵最小化的原理，采用改进的旗鱼优化器算法对变分模态分解的分解层数K和惩罚因子α寻优，克服了变分模态分解参数需人为设定的问题，提高了变分模态分解的自适应能力。变分模态分解联合小波阈值的滤波方法能有效过滤吸收光谱中的干扰噪声，更准确地读取二次谐波幅值，提高检测精度。
实施例2
参照图4-8，为本发明的一个实施例，提供了一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，为了验证本发明的有益效果，通过对比实验进行科学论证。
为模拟SF6电气设备内的气体混合状态，本次实验以SF6作为背景气体，以CO2作为目标检测气体组分，在实验室环境下采用SF6动态配气系统配置SF6和CO2的混合样气，配置的CO2样气体积分数包括0.87％、0.85％、0.83％、0.8％、0.77％、0.75％、0.72％、0.7％、0.68％、0.65％、0.62％、0.6％、0.57％、0.55％、0.53％、0.5％。将上述配置气体逐一充入可调谐吸收光谱技术检测装置进行实验，每组浓度采集60条二次谐波光谱数据，共采集得到960条光谱。
背景气体SF6和目标气体CO2在激光频谱范围内存在交叉干扰，同时可调谐吸收光谱技术检测装置采集到的信号包含大量高频噪声，因此，对原始数据进行滤波和背景扣除是光谱分析必不可少的步骤。
首先采用上文提出的自适应变分模态分解分解原始光谱，并根据每个分解模态与原始数据的相关系数选择相关性高的模态进行重构，最后采用小波阈值法对重构信号再次滤波，得到去除噪声后的吸收谱线。图4和图5分别是纯SF6和体积分数为0.85％的CO2吸收光谱滤波前后对比图，经过处理后的光谱信号不仅去除了原信号中的高频噪声，而且波形更平滑。
将纯SF6气体的吸收谱线作为背景信号，波长调制光谱法不同于直接吸收光谱法需要拟合背景基线，而是直接从目标气体的吸收谱线中减去背景信号。例如，将图4中的CO2吸收谱线减去图5中的SF6吸收谱线便可得到体积分数为0.85％的CO2扣除背景后的二次谐波吸收谱线。如图6所示，扣除背景后的二次谐波中心频率周围干扰谐波成分被有效削弱，能更容易准确读取到二次谐波的幅值。
图7是预处理过的16组不同浓度的CO2吸收谱线，CO2浓度越高二次谐波的幅值越高。根据最小二乘法线性拟合得到CO2的二次谐波峰值与浓度的线性线性相关系数R2＝0.988。
具体数值可见表1，表1浓度反演结果RMSE对比：

更进一步的，二次谐波幅值与被测气体浓度成正比，因此电力行业中最常用的浓度反演方法是利用最小二乘法建立二次谐波幅值与气体浓度的线性关系式。而本发明选择采用核极限学习机(KELM)来建立CO2的浓度预测模型，并利用改进的旗鱼优化器算法优化核极限学习机的正则化系数C及核参数S，适应度函数为预测浓度和真实浓度的均方根误差(RMSE)。
应说明的是，本次实验一共采集到960条吸收光谱，经过预处理后提取出960个二次谐波峰值，将二次谐波峰值和气体浓度分别作为所建模型的单输入和单输出。随机选取其中860对样本作为训练集用作模型训练，其余100对样本作为测试集用作模型精度测试。为验证本发明所提改进旗鱼优化器算法的有效性，将改进SFO-KELM与SFO-KELM、ELM(Sigmoidal激活函数，隐含层节点为20)、GA-BP(GA的交叉概率为0.7，变异概率为0.3；BP的训练次数为100，训练效率为0.01，隐含层节点为20)、PSO-KELM(PSO的惯性因子w为1，加速因子C1、C2均为2)、CLS进行模型精度对比。各算法迭代30次的实验结果如表1所示。
更进一步的，表1中ISFO-KELM的均方根误差(RMSE)最小，此时核极限学习机模型的正则化系数C＝1000，核参数S＝3.4177E-5。ISFO-KELM的训练集均方根误差比PSO-KELM小30.8％，比SFO-KELM小85.4％。ISFO-KELM的测试集均方根误差比PSO-KELM小36.9％，比SFO-KELM小86.8％。ISFO-KELM的训练集、测试集均方根误差均比GA-BP、ELM小2个数量等级，比线性最小二乘法(CLS)小3个数量等级。因此改进旗鱼优化器算法优化的核极限学习机在精度上均具有出色的表现。
应说明的是，以上实施例仅用以说明本发明的技术方案而非限制，尽管参照较佳实施例对本发明进行了详细说明，本领域的普通技术人员应当理解，可以对本发明的技术方案进行修改或者等同替换，而不脱离本发明技术方案的精神和范围，其均应涵盖在本发明的权利要求范围当中。
应说明的是，以上实施例仅用以说明本发明的技术方案而非限制，尽管参照较佳实施例对本发明进行了详细说明，本领域的普通技术人员应当理解，可以对本发明的技术方案进行修改或者等同替换，而不脱离本发明技术方案的精神和范围，其均应涵盖在本发明的权利要求范围当中。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。本申请实施例中的方案可以采用各种计算机语言实现，例如，面向对象的程序设计语言Java和直译式脚本语言JavaScript等。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
尽管已描述了本申请的优选实施例，但本领域内的技术人员一旦得知了基本创造性概念，则可对这些实施例作出另外的变更和修改。所以，所附权利要求意欲解释为包括优选实施例以及落入本申请范围的所有变更和修改。
显然，本领域的技术人员可以对本申请进行各种改动和变型而不脱离本申请的精神和范围。这样，倘若本申请的这些修改和变型属于本申请权利要求及其等同技术的范围之内，则本申请也意图包含这些改动和变型在内。

Claims (10)

  1. 一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，其特征在于：包括，
    基于传统旗鱼优化器，改进寻优精度和局部搜索能力，建立改进旗鱼优化器；
    收集原始吸收光谱，并结合所述改进旗鱼优化器优化的变分模态分解算法联合小波阈值法，对所述原始吸收光谱进行预处理；
    根据所述预处理结果，结合所述改进旗鱼优化器优化的核极限学习机建立吸收光谱的二次谐波幅值与CO2浓度的反演模型；
    根据所述反演模型进行CO2检测实验，判断SF6电气设备的运行状态。
  2. 如权利要求1所述的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，其特征在于：所述改进旗鱼优化器包括，
    设置算法控制参数，根据多策略初始化方式生成初始旗鱼和沙丁鱼种群；
    计算适应度值，记录本次迭代时的旗鱼和沙丁鱼全局最优位置；
    更新旗鱼位置，根据攻击力度更新沙丁鱼位置；
    分别对旗鱼和沙丁鱼的精英个体做柯西变异和自适应t分布变异；
    比较旗鱼和沙丁鱼最优解，根据结果替换旗鱼和沙丁鱼位置；
    迭代未结束则计算适应度值继续优化，迭代结束则返回最终结果。
  3. 如权利要求2所述的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，其特征在于：所述改进旗鱼优化器还包括，
    采用Tent混沌序列来改善旗鱼和沙丁鱼的初始种群分布，其数学表达式如下：
    其中，β是[0,1]的随机数。
  4. 如权利要求3所述的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，其特征在于：所述改进旗鱼优化器还包括，
    改进后的沙丁鱼位置更新数学表达式如下：

    Levy＝u/|v|1/β
    其中σ的数学表达式为：
    其中，u～N(0,σ2)，v～N(0,1)，β是[0,2]的随机数。
  5. 如权利要求4所述的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，其特征在于：所述改进旗鱼优化器还包括，对自适应变分模态分解和核极限学习机的参数进行寻优，对旗鱼和沙丁鱼的精英个体进行柯西变异和自适应t分布变异。
  6. 如权利要求5所述的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，其特征在于：所述反演模型包括，
    初始化改进旗鱼优化器基本参数，将自适应变分模态分解的多尺度排列熵之和作为适应度函数；
    生成初始种群，设置旗鱼种群和沙丁鱼种群比例；
    将各种群位置信息作为参数导入适应度函数计算多尺度排列熵，并将最小多尺度排列熵值和对应位置向量作为全局解和全局最优位置；
    更新旗鱼和沙丁鱼位置，重新计算适应度值，更新全局解和全局最优位置；进行优化迭代，直到达到最大迭代次数，将最后一次迭代时的全局最优位置作为自适应变分模态分解最佳参数，在此参数条件下进行自适应变分模态分解。
  7. 如权利要求6所述的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，其特征在于：所述反演模型还包括，将二次谐波峰值和气体浓度分别作为所建模型的单输入和单输出。
  8. 如权利要求7所述的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，其特征在于：所述反演模型还包括，采用核极限学习机来建立CO2的浓度预测模型，并利用改进旗鱼优化器算法优化核极限学习机的正则化系数及核参数，适应度函数为预测浓度和真实浓度的均方根误差。
  9. 如权利要求8所述的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，其特征在于：所述预处理包括，利用所述改进旗鱼优化器进行参数优化后的自适应变分模态分解，结合小波阈值法对原始吸收光谱进行预处 理。
  10. 如权利要求9所述的一种基于ISFO-VMD-KELM的SF6分解组分CO2浓度反演的方法，其特征在于：所述自适应变分模态分解包括，
    其中，f(t)是波形函数，即时间序列，t是时间，采集了1s的波形，f1(t)是幅值为1的40Hz余弦信号，f2(t)是幅值为3的150Hz间歇正弦信号，f3(t)是20dB的高斯白噪声。
PCT/CN2023/073975 2022-08-11 2023-01-31 一种基于isfo-vmd-kelm的sf6分解组分co2浓度反演的方法 WO2024031938A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210967017.9 2022-08-11
CN202210967017.9A CN115993345A (zh) 2022-08-11 2022-08-11 一种基于isfo-vmd-kelm的sf6分解组分co2浓度反演的方法

Publications (1)

Publication Number Publication Date
WO2024031938A1 true WO2024031938A1 (zh) 2024-02-15

Family

ID=85993112

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/073975 WO2024031938A1 (zh) 2022-08-11 2023-01-31 一种基于isfo-vmd-kelm的sf6分解组分co2浓度反演的方法

Country Status (2)

Country Link
CN (1) CN115993345A (zh)
WO (1) WO2024031938A1 (zh)

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180240018A1 (en) * 2016-05-19 2018-08-23 Jiangnan University Improved extreme learning machine method based on artificial bee colony optimization
CN111693481A (zh) * 2020-06-23 2020-09-22 贵州电网有限责任公司 测定sf6气体中co含量非分散红外吸收光谱标定方法
CN112364758A (zh) * 2020-11-10 2021-02-12 湖北惠立网络科技有限公司 基于多目标图像识别的垃圾分类回收方法及系统
WO2021056727A1 (zh) * 2019-09-27 2021-04-01 山东科技大学 一种基于变分模态分解和排列熵的联合降噪方法
CN113241087A (zh) * 2021-05-10 2021-08-10 青岛科技大学 基于改进vmd和改进小波阈值法相结合的水声信号去噪方法
CN114330194A (zh) * 2021-12-31 2022-04-12 桂林电子科技大学 基于sfo优化深度极限学习机的模拟电路故障诊断方法
CN114565237A (zh) * 2022-02-15 2022-05-31 石河子大学 用于多能组合发电站的综合能源低碳调度方法及系统
CN114611411A (zh) * 2022-03-31 2022-06-10 河北工业大学 基于isfo-svm模型的igbt结温预测方法
US20220197233A1 (en) * 2020-12-18 2022-06-23 Wuhan University Wind power prediction method and system for optimizing deep transformer network

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180240018A1 (en) * 2016-05-19 2018-08-23 Jiangnan University Improved extreme learning machine method based on artificial bee colony optimization
WO2021056727A1 (zh) * 2019-09-27 2021-04-01 山东科技大学 一种基于变分模态分解和排列熵的联合降噪方法
CN111693481A (zh) * 2020-06-23 2020-09-22 贵州电网有限责任公司 测定sf6气体中co含量非分散红外吸收光谱标定方法
CN112364758A (zh) * 2020-11-10 2021-02-12 湖北惠立网络科技有限公司 基于多目标图像识别的垃圾分类回收方法及系统
US20220197233A1 (en) * 2020-12-18 2022-06-23 Wuhan University Wind power prediction method and system for optimizing deep transformer network
CN113241087A (zh) * 2021-05-10 2021-08-10 青岛科技大学 基于改进vmd和改进小波阈值法相结合的水声信号去噪方法
CN114330194A (zh) * 2021-12-31 2022-04-12 桂林电子科技大学 基于sfo优化深度极限学习机的模拟电路故障诊断方法
CN114565237A (zh) * 2022-02-15 2022-05-31 石河子大学 用于多能组合发电站的综合能源低碳调度方法及系统
CN114611411A (zh) * 2022-03-31 2022-06-10 河北工业大学 基于isfo-svm模型的igbt结温预测方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
HONGLIANG CHEN, LI YANG;ZHANG YIHUI;XU FEIYUN: "Crane fault diagnosis based on adaptive optimization variational modal decomposition", HOISTING AND CONVEYING MACHINERY, vol. 9, 15 May 2022 (2022-05-15), pages 49 - 56, XP093138128 *
LI GUO-LIN, JIAO YUE, MA KUN, YAN ZHI-DAN, ZHANG XUE-NA, WU YUN-HUI, ZHANG ZE-CHENG: "Application of near-infrared CO<sub>2</sub> concentration inversion system based on empirical modal decomposition", OPTICS AND PRECISION ENGINEERING, ZHONGGUO KEXUEYUAN CHANGCHUN JINGMI JIXIE YANJIUSUO, CHANGCHUN, CN, vol. 29, no. 5, 15 May 2021 (2021-05-15), CN , pages 940 - 950, XP093138127, ISSN: 1004-924X, DOI: 10.37188/OPE.20212905.0940 *

Also Published As

Publication number Publication date
CN115993345A (zh) 2023-04-21

Similar Documents

Publication Publication Date Title
He et al. Baseline correction for Raman spectra using an improved asymmetric least squares method
Liu et al. SEAGLE: Sparsity-driven image reconstruction under multiple scattering
Meng et al. A modified empirical mode decomposition algorithm in TDLAS for gas detection
Chen et al. An adaptive denoising method for Raman spectroscopy based on lifting wavelet transform
D’Abramo et al. Theoretical modeling of UV-Vis absorption and emission spectra in liquid state systems including vibrational and conformational effects: Explicit treatment of the vibronic transitions
WO2014137564A1 (en) Systems and methods for boosting coal quality measurement statement of related cases
EP3120243A1 (fr) Procede et dispositif d&#39;aide a l&#39;optimisation et la parallelisation de code
Trewartha et al. Quark propagation in the instantons of lattice QCD
CN104345049B (zh) 激光诱导击穿光谱的小波阈值降噪的阈值校正方法
Jiang et al. Qualitative and quantitative analysis in solid-state fermentation of protein feed by FT-NIR spectroscopy integrated with multivariate data analysis
Tian et al. Near-infrared methane sensor with neural network filtering
Liu et al. Continuous background correction using effective points selected in third-order minima segments in low-cost laser-induced breakdown spectroscopy without intensified CCD
Ruan et al. A novel hybrid filter/wrapper method for feature selection in archaeological ceramics classification by laser-induced breakdown spectroscopy
Kistenev et al. Super-resolution reconstruction of noisy gas-mixture absorption spectra using deep learning
Lian et al. Filter pruning via separation of sparsity search and model training
WO2024031938A1 (zh) 一种基于isfo-vmd-kelm的sf6分解组分co2浓度反演的方法
Zhang et al. Denoising method based on CNN-LSTM and CEEMD for LDV signals from accelerometer shock testing
Kazemzadeh et al. Deep learning as an improved method of preprocessing biomedical Raman spectroscopy data
Tekawade et al. 3d autoencoders for feature extraction in X-ray tomography
CN113297722A (zh) 滨海土壤盐分评估方法及系统
Lukić et al. Neural networks-based real-time determination of the laser beam spatial profile and vibrational-to-translational relaxation time within pulsed photoacoustics
CN115950827A (zh) 一种降低开放光路光谱检测信号噪声影响的方法
CN114062306B (zh) 一种近红外光谱数据分段预处理方法
CN113686810B (zh) 一种基于卷积神经网络的近红外光谱波长选择方法
Bleicker et al. Strong quenches in the one-dimensional Fermi-Hubbard model

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23851174

Country of ref document: EP

Kind code of ref document: A1