Soil organic matter content inversion method based on near infrared spectrum

Abstract

The invention discloses a soil organic matter content inversion method based on a near infrared spectrum, and relates to the technical field of soil detection. Compared with a previous soil organic matter content measurement technology, the problems that in the aspect of soil organic matter content measurement, an existing near infrared spectrum technology is complex in data preprocessing, difficult in feature selection and insufficient in model generalization ability are solved; the near infrared spectrum range is optimally selected, key characteristic wave bands related to soil organic matter changes are effectively focused, and the accuracy and reliability of measurement are improved; a differential transformation preprocessing technology is used, and the data quality is effectively improved; efficient feature selection is carried out by adopting a stable competitive self-adaptive reweighted sampling method, a feature wave band which can reflect the soil organic matter content is extracted, and the prediction capability of the model is enhanced. And a plurality of machine learning algorithms and parameter optimization technologies are combined to optimize the model structure and parameters so as to adapt to soil samples of different areas and different types, and the prediction precision and stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of soil detection, and in particular to a soil organic matter content inversion method based on near infrared spectroscopy. Background Art

[0002] As an important indicator of soil fertility, changes in soil organic matter content directly affect the content of water and other microorganisms in the soil, and thus have an important impact on the biological, chemical and physical properties of the soil. Therefore, quickly and accurately obtaining the content of soil organic matter has become an inevitable requirement for the development of modern agriculture, and plays a vital role in maintaining ecological balance and promoting sustainable agricultural development. In the past few decades, the measurement technology of soil organic matter content has been widely studied by researchers, which can be roughly divided into traditional chemical methods and hyperspectral technology. The determination of soil organic matter content using traditional chemical methods has high accuracy and is more stable and reliable, but it also has many disadvantages such as cumbersome operation process, environmental pollution, high cost, time-consuming and labor-intensive, and the need for measurement personnel to have rich experience. Compared with traditional chemical methods, hyperspectral technology has a simpler operation method, less resource loss, faster analysis speed and no pollution to the environment while maintaining high accuracy. It uses soil spectral reflectance to reflect the optical properties of soil at different wavelengths. By measuring and analyzing the soil spectral reflectance, the physical and chemical properties and characteristics of the soil are obtained.

[0003] At present, many scholars have established inversion models based on hyperspectral technology by combining different spectral processing methods with different machine learning prediction models to monitor the soil organic matter content in different regions.

[0004] Spectral data preprocessing algorithm: Soil structure and composition are complex, and its spectrum is obtained by the comprehensive superposition of the reflectance of various components. The spectral information is often mixed with feedback information of various factors such as salt and moisture, which will cause the soil spectrum to be non-specific, resulting in serious impact and deviation on the accuracy of measuring soil organic matter through spectral data. In addition, due to the influence of factors such as the acquisition environment and the noise of the instrument itself, the high-spectrum obtained in the laboratory often contains some noise. At present, domestic and foreign scholars have carried out a large number of estimation studies on spectral preprocessing and denoising methods. Among them, a large number of research methods show that preprocessing methods such as spectral denoising, spectral transformation methods and spectral data feature band selection can effectively improve the reliability and accuracy of spectral data. Commonly used denoising methods include moving average method, Savitzky-Golay smoothing, etc.; spectral transformation methods include different forms of differential transformation, wavelet transformation, etc.; feature selection methods include principal component analysis, correlation analysis, etc.

[0005] Inversion model: Since the birth of hyperspectral remote sensing technology, a large number of experts and scholars at home and abroad have conducted in-depth research on soil modeling inversion methods. Today, modeling methods are constantly being improved and the accuracy of the models is also constantly improving. At present, the commonly used modeling inversion methods include multivariate stepwise regression, partial least squares regression, support vector machine, BP neural network, random forest, convolutional neural network, etc. These studies have shown that near-infrared spectroscopy technology can significantly improve the efficiency of soil organic matter content measurement while maintaining a high degree of accuracy.

[0006] Although the existing near-infrared spectroscopy technology has made some progress in measuring soil organic matter content, it still has the following shortcomings:

[0007] Complex data preprocessing: Some commonly used methods, such as wavelet transform and Fourier transform, can improve data quality to a certain extent, but their effects are limited when faced with complex soil spectral data. In contrast, differential transform is particularly important in data preprocessing because it can effectively highlight the detailed changes in spectral data. Differential transform can not only enhance the detailed characteristics of spectral signals, but also effectively reduce the impact of background noise and improve the accuracy of subsequent analysis.

[0008] Difficulty in feature selection: Currently, most spectral studies on soil are concentrated in the bands from visible light to short-wave infrared (400nm-2500nm). Soil spectral data contains a lot of redundant information. How to select the characteristic bands that best reflect the content of soil organic matter is a difficult problem. Existing methods such as correlation analysis, principal component analysis (PCA), and stepwise regression have been applied to the selection of characteristic bands. Although these methods can reduce the data dimension and extract the main features, they still have shortcomings in extracting the optimal features. For example, correlation analysis may miss some important nonlinear features, and PCA may lose important information while retaining the main features. These methods have limited performance in capturing spectral features related to soil organic matter, and it is difficult to fully reflect the complex characteristics of soil organic matter.

[0009] Limitations of inversion models: Models are diverse but difficult to choose. Currently, there are many types of models used to predict soil organic matter content, such as multiple linear regression (MLR), partial least squares regression (PLSR), support vector machine (SVM), etc. These models have their own advantages and disadvantages, and perform differently in different data sets and application scenarios. For example, MLR and PLSR work well when dealing with linear relationships, but have limited capabilities when facing complex nonlinear relationships; although SVM has advantages in capturing nonlinear characteristics, it has high requirements for parameter selection and model training, and is easily affected by sample size and quality. There are still deficiencies in the selection of models and the optimization of their parameters, especially in practical applications, the complexity of soil samples in different regions or different types increases the difficulty of model generalization. The generalization ability of existing models is limited, and it is difficult to maintain consistent prediction results under different environments and conditions. The inadequacy of model selection and parameter optimization makes it difficult to achieve ideal levels of prediction accuracy and stability in practical applications.

[0010] In order to solve the above problems, the present invention proposes a soil organic matter content inversion method based on near infrared spectroscopy. Summary of the invention

[0011] The purpose of the present invention is to propose a soil organic matter content inversion method based on near infrared spectroscopy to solve the problems raised in the background technology:

[0012] The existing near-infrared spectroscopy technology still has problems in measuring soil organic matter content, such as complex data preprocessing, difficult feature selection and insufficient model generalization ability.

[0013] In order to achieve the above object, the present invention adopts the following technical solutions:

[0014] A soil organic matter content inversion method based on near infrared spectroscopy includes the following steps:

[0015] S1: Set up soil sampling points and collection plans to collect different types of soil samples;

[0016] S2: Biochemical experiments are conducted on the sampled soil samples to obtain the measured value of soil organic matter; soil spectral data of soil samples are collected through a hyperspectral camera, and pre-processed and characteristic wavelength selected;

[0017] S3: Establish a prediction model for soil organic matter content based on the marine predator algorithm and random forest model and evaluate and optimize the model;

[0018] S4: Data testing based on the optimized prediction model of soil organic matter content.

[0019] Preferably, the types of soil samples collected in S1 include red soil, paddy soil and tidal soil.

[0020] Preferably, in S2, soil spectral reflectance is measured based on a hyperspectral camera with a spectral range of 900nm to 1700nm.

[0021] Preferably, in the preprocessing step of S2, biochemical experiments are carried out on the sampled soil samples to obtain the measured value of soil organic matter; the original spectrum is denoised based on 5G smoothing, and the soil spectral data is further preprocessed using several differential transformation methods.

[0022] Preferably, the differential transformation method includes first-order differential, second-order differential, logarithmic first-order differential, logarithmic second-order differential, reciprocal first-order differential, reciprocal second-order differential, logarithmic reciprocal first-order differential, logarithmic reciprocal second-order differential, reciprocal logarithmic first-order differential, and reciprocal logarithmic second-order differential.

[0023] Preferably, the characteristic wavelength selection processing in S2 is based on the stable competitive adaptive reweighted sampling method to find the optimal variable subset of spectral data, specifically as follows: the stable competitive adaptive reweighted sampling method calculates the stability of each wavelength variable, and the first p variables with larger stability values ​​are screened out as a new variable subset through the stable competitive adaptive reweighted sampling method and the exponential decay function; after multiple cycles of screening, the final variable subset is obtained, and the final variable subset is calculated through the partial least squares regression model to obtain its cross-validation root mean square error; finally, the subset with the smallest error value is selected as the optimal variable subset, that is, the optimal characteristic band.

[0024] Preferably, in S3, a nonlinear relationship between organic matter content and natural environmental variables and socioeconomic factors is simulated based on a machine learning algorithm to build a prediction model for soil organic matter content; wherein the structure and parameters of the machine learning algorithm are optimized based on the MPA algorithm; the MPA algorithm is specifically as follows:

[0025] MPA starts the optimization process by initializing the search space for the prey position X0 as follows:

[0026] X0=X min +rand(X max -X min )

[0027] Among them, X max , X min is the search space range, rand(·) is a random number between (0,1);

[0028] The optimization phase of MPA is divided into three cases, as follows:

[0029] The first is when the speed ratio is high or the prey moves faster than the predator. This happens at the beginning of the optimization iteration. At this stage, the best strategy for the predator is to maintain the current position. The optimization process based on the exploration strategy is as follows:

[0030]

[0031] Among them, i=1, Λ, n is the population size; stepsize1 is the moving step size of this stage; R B is a normally distributed Brownian walk random vector; is a term-by-term multiplication operator used to simulate the movement of prey; Prey′1 is the prey matrix updated in this stage; Prey is the prey matrix; P = 0.5; R is a random uniform distribution value of [0,1]; Elite is the elite matrix composed of top predators; Iter is the current iteration number; MaxIter is the maximum iteration number;

[0032] The second is the unit speed ratio or when the predator and prey move at equal speeds, which occurs in the middle of the optimization iteration. The population is divided into two parts, where the prey does Levy motion, responsible for the algorithm to develop in the search space, and the predator does Brownian motion, responsible for the algorithm to explore in the search space. The optimization process of the development and exploration strategies is as follows:

[0033]

[0034] Among them, R L is the random vector of Levy motion; CF is the adaptive parameter of the predator's moving step size; stepsize 21 For this stage, the moving step length of the population of the Levy motion part; Prey' 21 The prey matrix for the population update of the Levy motion part for this stage; stepsize 22 The moving step length of the Brownian motion population for this stage; Prey' 22 The prey matrix for the Brownian motion population update for this stage;

[0035] The third is a low speed ratio or when the predator moves faster than the prey, which occurs in the late iteration of the optimization. At this time, the best strategy for the predator is Levy motion. The optimization process of the exploration strategy is as follows:

[0036]

[0037] Among them, stepsize 31 The moving step length for this phase; Prey' 31 The prey matrix updated for this phase;

[0038] There are also eddy current and fish aggregation effect strategies in the exploration process; the eddy current and fish aggregation effect strategies are as follows:

[0039]

[0040] Among them, FADs = 0.2; r is a random number between [0,1]; U is a binary array with d dimensions, and the value of each dimension is determined according to the relationship between the random number r and the size of FADs; n1,n2∈[1,n] are random subscripts of Prey.

[0041] Compared with the prior art, the present invention provides a soil organic matter content inversion method based on near infrared spectroscopy, which has the following beneficial effects:

[0042] The present invention optimizes the near-infrared spectral range, effectively focuses on the key characteristic bands related to soil organic matter changes, reduces redundant information, improves the signal-to-noise ratio, and thus improves the accuracy and reliability of the measurement; uses differential transformation preprocessing technology to effectively remove noise and interference factors, highlight the detail changes in the spectral data, improve data quality, and provide a reliable basis for subsequent analysis; adopts an efficient feature selection algorithm, stable competitive adaptive reweighted sampling, extracts the characteristic bands that best reflect the soil organic matter content from a large amount of spectral data, and enhances the prediction ability of the model. Combined with a variety of machine learning algorithms and parameter optimization techniques, the model structure and parameters are optimized, so that the model can adapt to soil samples in different regions and different types, and improve prediction accuracy and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of the method mentioned in Example 1 of the present invention;

[0044] Figure 2 This is a schematic diagram of the spectrum equipment acquisition device mentioned in Example 1 of the present invention;

[0045] Figure 3 This is a flow chart of the SCARS algorithm mentioned in Example 1 of the present invention. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0047] The present invention optimizes the near-infrared spectral range, effectively focuses on the key characteristic bands related to soil organic matter changes, reduces redundant information, improves the signal-to-noise ratio, and thus improves the accuracy and reliability of the measurement; uses differential transformation preprocessing technology to effectively remove noise and interference factors, highlight the detailed changes in the spectral data, improve data quality, and provide a reliable basis for subsequent analysis; adopts an efficient feature selection algorithm, stable competitive adaptive reweighted sampling, extracts the characteristic bands that best reflect the soil organic matter content from a large amount of spectral data, and enhances the prediction ability of the model. Combined with a variety of machine learning algorithms and parameter optimization techniques, the model structure and parameters are optimized so that the model can adapt to soil samples from different regions and different types, and improve prediction accuracy and stability. Specifically include the following contents.

[0048] Embodiment 1:

[0049] See also Figure 1-3 The present invention provides a soil organic matter content inversion method based on near infrared spectroscopy, comprising the following steps:

[0050] S1: Set up soil sampling points and collection plans to collect different types of soil samples; the details are as follows:

[0051] A total of 85 soil sample points were collected on different types of 0-20cm surface soil in the study area, and 5 surface soil mixed soil samples were taken at each sampling point. The soil samples are generally red soil, paddy soil or tidal soil. S2: Biochemical experiments were conducted on the sampled soil samples to obtain the measured values ​​of soil organic matter; soil spectral data of soil samples were also collected through a hyperspectral camera, and preprocessed and characteristic wavelength selected; the details are as follows:

[0052] The mixed soil samples were brought back to the room for natural air drying, ground to remove impurities and screened through a 2 mm sieve. The soil samples were divided into two parts, one for soil spectral information collection and the other for chemical method determination of organic matter content. The chemical method for determining soil SOM content used potassium dichromate oxidation method. The soil spectral reflectance was measured using the American FX (17) hyperspectral camera, which has a spectral range of 900nm-1700nm, a spectral resolution of 3.5nm, and a total of 224 bands. The spectral equipment acquisition device is as follows: Figure 2 As shown in the figure, during the measurement, six 50W halogen lamps provided the light source, the field of view of the fiber optic probe was 10°, and the distance from the soil sample surface was 20cm. Each soil sample was divided into 27 parts and placed in a tray. Before each sample was measured, a white board calibration was performed to reduce the error caused by the instrument. Then, based on the ENVI5.3 software, 27 spectral curves were measured for each soil sample, and the average value was taken as the actual spectral curve of the soil sample after removing the outliers.

[0053] Soil spectral data preprocessing can reduce the influence of factors such as noise, stray signals and nonlinear spectral effects in soil spectral data, and improve the accuracy and reliability of data. This preprocessing method can perform non-destructive analysis on soil samples, and can also provide more efficient, accurate and comprehensive data information for soil analysis, thereby providing important support and guarantee for soil science research. This embodiment first uses SG smoothing to denoise the original spectrum, and further uses a variety of differential transformation methods to preprocess the soil spectral data to further improve the correlation between reflectance data and soil indicators. Ten differential transformation methods were collected and sorted out, including first-order differential (FD), second-order differential (SD), logarithmic first-order differential (LFD), logarithmic second-order differential (LSD), reciprocal first-order differential (RFD), reciprocal second-order differential (RSD), log reciprocal first-order differential (LRFD), log reciprocal second-order differential (LRSD), reciprocal logarithmic first-order differential (RLFD), and reciprocal logarithmic second-order differential (RLSD). The theoretical formulas of the above differential transformation methods are shown in Table 1, where log is based on e, which represents the spectral data after SG smoothing, and x' and x" represent the first-order differential and second-order differential, respectively.

[0054] Table 1 Differential transformation method and its theoretical formula

[0055]

[0056] SCARS has obvious advantages in solving high-dimensional data and complex problems. It can better capture the intrinsic structure of data and the relationship between features during feature selection and sample resampling. It can not only automatically adjust the weight of samples according to the distribution of data, thereby effectively processing unbalanced or noisy data sets, but also quickly converge to the global optimal solution and retain the information of the original data to the greatest extent while maintaining a small sample set size, thereby improving the prediction performance and generalization ability of the model.

[0057] The calculation process of finding the optimal variable subset of spectral data using the stable competitive adaptive reweighted sampling method includes n cycles, and its flow chart is as follows: Figure 3 As shown. In each cycle, the stable competitive adaptive reweighted sampling method first calculates the stability of each wavelength variable, and then selects the variables with larger stability values ​​as new variable subsets through the stable competitive adaptive reweighted sampling method (SCARS) and the exponential decay function (EDF). After multiple cycles of screening, the final variable subset is obtained, and the final variable subset is calculated through the partial least squares regression model (PLS) to obtain its cross-validation root mean square error (RMSECV). Finally, the subset with the smallest RMSECV value is selected as the optimal variable subset, that is, the optimal characteristic band.

[0058] S3: Based on the marine predator algorithm and random forest, a prediction model for soil organic matter content was established and the model was evaluated and optimized; the details are as follows:

[0059] The prediction model of soil organic matter content is built based on machine learning algorithms, such as PLSR, SVR, RFB and PNN. Taking the RF model as an example, the random forest model (RF) constructs several decision trees to train the data to simulate the nonlinear relationship between organic matter content and natural environmental variables and socioeconomic factors; RF is an integrated learning algorithm based on decision trees. By constructing multiple decision trees, each decision tree independently predicts the data. RF can better fit high-dimensional data and has a certain ability to resist overfitting. In RF regression, the construction of each decision tree is random to ensure the diversity and generalization ability of the model. The RF model is constructed through the built-in function in MATLAB software, and the model parameters are further optimized by calculating the feature importance. The tree parameter is set to 100 according to the soil sample data.

[0060] The structure and parameters of the random forest model are optimized based on the MPA algorithm; the MPA algorithm is as follows:

[0061] MPA starts the optimization process by initializing the search space for the prey position X0 as follows:

[0062] X0=X min +rand(X max -X min )

[0063] Among them, X max , X min is the search space range, rand(·) is a random number between (0,1);

[0064] The optimization phase of MPA is divided into three cases, considering different speed ratios, while simulating the entire life cycle of predators and prey. The details are as follows:

[0065] The first is when the speed ratio is high or the prey moves faster than the predator. This happens in the early stages of the optimization iteration, when global search is very important. The best strategy for the predator at this stage is to maintain the current position. The optimization process based on the exploration strategy is as follows:

[0066]

[0067] Among them, i=1, Λ, n is the population size; stepsize1 is the moving step size of this stage; R B is a normally distributed Brownian walk random vector; is a term-by-term multiplication operator used to simulate the movement of prey; Prey′1 is the prey matrix updated in this stage; Prey is the prey matrix; P = 0.5; R is a random uniform distribution value of [0,1]; Elite is the elite matrix composed of top predators; Iter is the current iteration number; MaxIter is the maximum iteration number;

[0068] The second is the unit speed ratio or when the predator and prey move at almost the same speed. This occurs in the middle of the optimization iteration. The population is divided into two parts, where the prey performs Levy motion, responsible for the algorithm to develop in the search space, and the predator performs Brownian motion, responsible for the algorithm to explore in the search space. The optimization process of the development and exploration strategies is as follows:

[0069]

[0070] Among them, R L is the random vector of Levy motion; CF is the adaptive parameter of the predator's moving step size; stepsize 21 The moving step length of the population of the Lévy motion for this stage; Prey′ 21 The prey matrix for the population update of the Levy motion part for this stage; stepsize 22 The moving step length of the Brownian motion population for this stage; Prey′ 22 The prey matrix for the Brownian motion population update for this stage;

[0071] The third is a low speed ratio or when the predator moves faster than the prey, which occurs in the late iteration of the optimization. At this time, the best strategy for the predator is Levy motion. The optimization process of the exploration strategy is as follows:

[0072]

[0073] Among them, stepsize 31 is the moving step length for this stage; Prey′ 31 The prey matrix updated for this phase;

[0074] There are also eddy current and fish aggregation effect strategies in the exploration process; this strategy enables MPA to overcome the early convergence problem in the optimization process and avoid local extreme values ​​in the optimization process. The details are as follows:

[0075]

[0076] Among them, FADs = 0.2; r is a random number between [0,1]; U is a binary array with d dimensions, and the value of each dimension is determined according to the relationship between the random number r and the size of FADs; n1,n2∈[1,n] are random subscripts of Prey.

[0077] Since the number of generated trees (ntree) and the maximum depth of the tree (mdepth) in the parameters of random forest have a great impact on the accuracy of the model, this embodiment uses the MPA algorithm to optimize the parameters ntree and mdepth of the random forest, and uses the optimal parameter values ​​to establish the MPA-RF model. The specific steps of the MPA-RF algorithm are as follows: set the population size of the MPA algorithm to 20, the maximum number of iterations to 200, and set the values ​​of parameters such as FDAs, and compare the current number of iterations with the maximum number of iterations to control the operation of the algorithm. 2 , RMSE, RPD and other indicators are used to evaluate and optimize the model.

[0078] S4: Data testing based on the optimized prediction model of soil organic matter content.

[0079] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A soil organic matter content inversion method based on near infrared spectroscopy, characterized in that: The steps include: S1: Set up soil sampling points and collection plans to collect different types of soil samples; S2: Biochemical experiments are conducted on the sampled soil samples to obtain the measured value of soil organic matter; soil spectral data of soil samples are collected through a hyperspectral camera, and pre-processed and characteristic wavelength selected; S3: Establish a prediction model for soil organic matter content based on the marine predator algorithm and random forest model and evaluate and optimize the model; S4: Data testing based on the optimized prediction model of soil organic matter content.

2. The soil organic matter content inversion method based on near infrared spectroscopy according to claim 1 is characterized in that: The types of soil samples collected in S1 include red soil, paddy soil and fluvo-aquic soil.

3. The soil organic matter content inversion method based on near infrared spectroscopy according to claim 1 is characterized in that: In S2, soil spectral reflectance is measured based on a hyperspectral camera with a spectral range of 900nm to 1700nm.

4. The soil organic matter content inversion method based on near infrared spectroscopy according to claim 1 is characterized in that: In the preprocessing step of S2, biochemical experiments are carried out on the sampled soil samples to obtain the measured value of soil organic matter; the original spectrum is denoised based on 5G smoothing, and the soil spectrum data is further preprocessed using several differential transformation methods.

5. The soil organic matter content inversion method based on near infrared spectroscopy according to claim 4 is characterized in that: The differential transformation methods include first-order differential, second-order differential, logarithmic first-order differential, logarithmic second-order differential, reciprocal first-order differential, reciprocal second-order differential, logarithmic reciprocal first-order differential, logarithmic reciprocal second-order differential, reciprocal logarithmic first-order differential, and reciprocal logarithmic second-order differential.

6. The soil organic matter content inversion method based on near infrared spectroscopy according to claim 1 is characterized in that: The characteristic wavelength selection process in S2 is based on the stable competitive adaptive reweighted sampling method to find the optimal variable subset of the spectral data, which is as follows: the stable competitive adaptive reweighted sampling method calculates the stability of each wavelength variable, and the variables with the largest stability values ​​of the first p are selected as the new variable subset through the stable competitive adaptive reweighted sampling method and the exponential decay function; After multiple rounds of screening, the final variable subset is obtained, and the final variable subset is calculated through the partial least squares regression model to obtain its cross-validation root mean square error; finally, the subset with the smallest error value is selected as the optimal variable subset, that is, the optimal characteristic band.

7. The soil organic matter content inversion method based on near infrared spectroscopy according to claim 1 is characterized in that: In S3, the nonlinear relationship between organic matter content and natural environmental variables and socioeconomic factors is simulated based on a machine learning algorithm to build a prediction model for soil organic matter content; wherein the structure and parameters of the machine learning algorithm are optimized based on the MPA algorithm; the MPA algorithm is specifically as follows: MPA starts the optimization process by initializing the search space for the prey position X0 as follows: X0=X min +rand(X max -X min ) Among them, X max , X min is the search space range, rand(·) is a random number between (0,1); The optimization phase of MPA is divided into three cases, as follows: The first is when the speed ratio is high or the prey moves faster than the predator. This happens at the beginning of the optimization iteration. At this stage, the best strategy for the predator is to maintain the current position. The optimization process based on the exploration strategy is as follows: Among them, i=1, Λ, n is the population size; stepsize1 is the moving step size of this stage; R B is a normally distributed Brownian walk random vector; is a term-by-term multiplication operator used to simulate the movement of prey; Prey1' is the prey matrix updated in this stage; Prey is the prey matrix; P = 0.5; R is a random uniform distribution value of [0,1]; Elite is the elite matrix composed of top predators; Iter is the current number of iterations; MaxIter is the maximum number of iterations; The second is the unit speed ratio or when the predator and prey move at equal speeds, which occurs in the middle of the optimization iteration. The population is divided into two parts, where the prey does Levy motion, responsible for the algorithm to develop in the search space, and the predator does Brownian motion, responsible for the algorithm to explore in the search space. The optimization process of the development and exploration strategies is as follows: Among them, R L is the random vector of Levy motion; CF is the adaptive parameter of the predator's moving step size; stepsize 21 For this stage, the moving step length of the population of the Levy motion part; Prey' 21 The prey matrix for the population update of the Levy motion part for this stage; stepsize 22 The moving step length of the Brownian motion population for this stage; Prey' 22 The prey matrix for the Brownian motion population update for this stage; The third is a low speed ratio or when the predator moves faster than the prey, which occurs in the late iteration of the optimization. At this time, the best strategy for the predator is Levy motion. The optimization process of the exploration strategy is as follows: Among them, stepsize 31 The moving step length for this phase; Prey' 31 The prey matrix updated for this phase; There are also eddy current and fish aggregation effect strategies in the exploration process; the eddy current and fish aggregation effect strategies are as follows: Among them, FADs = 0.2; r is a random number between [0,1]; U is a binary array with d dimensions, and the value of each dimension is determined according to the relationship between the random number r and the size of FADs; n1,n2∈[1,n] are random subscripts of Prey.

Images