Ground temperature field change prediction method based on fuzzy least square support vector machine

Abstract

The invention provides a ground temperature field change prediction method based on a fuzzy least square support vector machine, and belongs to the technical field of ground temperature field prediction, and the method comprises the following steps: S1, based on a drill hole distribution diagram, arranging shallow ground temperature monitoring points, and obtaining initial monitoring parameters; s2, performing Pearson correlation analysis based on the initial parameters, and determining input parameters of the prediction model; s3, performing fuzzification processing on the input parameters, and establishing a fuzzy least square support vector machine model; s4, parameters are optimized based on a fruit fly optimization algorithm, and self-training and self-verification of a fuzzy least square support vector machine model are carried out; and S5, performing ground temperature field prediction based on the trained and verified fuzzy least square support vector machine model. According to the method, fuzzy processing is carried out on the data, subsequent effective processing of the data is facilitated, and the rapidity and accuracy of model prediction are effectively improved by fully utilizing the least square support vector machine.

Description

Technical Field

[0001] The invention belongs to the technical field of geothermal field prediction, and in particular relates to a geothermal field change prediction method based on fuzzy least squares support vector machine. Background Art

[0002] The operation of the ground source heat pump air conditioning system will cause changes in the ground temperature field, especially in areas where the cold and hot loads are unbalanced throughout the year. On the one hand, the change in the ground temperature field will affect the outlet water temperature of the buried pipe, thereby changing the operating performance of the heat pump unit; on the other hand, the change in the ground temperature field will damage the underground ecological environment to a certain extent. Therefore, it is of great significance to maintain the continuous stability of the ground temperature field. The real-time operation of the ground source heat pump system will inevitably cause instantaneous changes in the ground temperature field. In order to restore the ground temperature in time and suppress the further expansion of the ground temperature field change, it is necessary to grasp the change of the ground temperature field in real time.

[0003] Numerical simulation is a commonly used method for predicting geothermal fields. However, due to the complex underground geological structure, it is difficult to obtain complete underground rock and soil information. Therefore, the model obtained using the physical modeling method cannot guarantee its accuracy. The prediction of geothermal fields is achieved by arranging sensors through drilling. This method has high acquisition costs and underground sensors are difficult to repair after being damaged. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, a method for predicting geothermal field changes based on fuzzy least squares support vector machine is provided.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] This technical solution proposes a method for predicting geothermal field changes based on fuzzy least squares support vector machine, which includes the following steps:

[0007] S1: Based on the borehole distribution map, arrange shallow geothermal monitoring points and obtain initial monitoring parameters;

[0008] S2: Based on the initial parameters, Pearson correlation analysis is performed to determine the input parameters of the prediction model;

[0009] S3: Fuzzify the input parameters and establish a fuzzy least squares support vector machine model;

[0010] S4: Optimize the parameters based on the fruit fly optimization algorithm, and perform self-training and self-validation of the fuzzy least squares support vector machine model;

[0011] S5: Predict the geothermal field based on the trained and verified fuzzy least squares support vector machine model.

[0012] Preferably, in S1, the initial parameters include the supply and return water temperature on the user side, the supply and return water temperature, flow rate, and ambient temperature on the ground source side.

[0013] Preferably, in S2, the Pearson correlation coefficient is as follows:

[0014]

[0015] In the formula, represents the average value of x, represents the mean value of y;

[0016] Determine r xy Parameters whose absolute values ​​are not less than 0.5 are the input parameters of the prediction model.

[0017] Preferably, in S3, a Gaussian membership function is used to perform fuzzy processing on the input parameters, and the formula is as follows:

[0018]

[0019] In the formula, k represents the number of fuzzy sets, T represents the input sample, T0(i) and σ represent the symmetry axis and variance of the i-th membership function, respectively. Represents the fuzzy membership of the jth input data of sample i.

[0020] Preferably, the fuzzy least squares support vector machine model establishment method includes:

[0021] S11: Define the training set as:

[0022] T={(x i ,y i )|x i ∈R m ,y i ∈R n ,i=1,2,…l} (3);

[0023] In the formula, l represents the total number of samples, R m represents the input space, R n Represents the output space;

[0024] The decision function is defined as:

[0025]

[0026] Where w∈R m represents the normal vector, b represents the threshold, Represents the mapping of input data in high-dimensional space;

[0027] S12: Transformed into solving the constrained minimum problem, the formula is as follows:

[0028]

[0029] Where J represents structural risk, y i represents the constraint condition, c represents the penalty parameter, e i represents the error variable;

[0030] S13: Introduce Lagrange multiplier α=[α1,α2,…α n ], construct the Lagrangian function, the formula is as follows:

[0031]

[0032] S14: Solve partial derivatives of w, b, e, and α respectively, and eliminate w and e i , the formula is as follows:

[0033]

[0034] Define the radial basis function as the kernel function, the formula is as follows:

[0035]

[0036] In the formula, σ is used as the kernel function width adjustment parameter to control the range of the radial basis function;

[0037] S15: Combine equations (7) and (8) to obtain the matrix of the linear equation system, as follows:

[0038]

[0039] Solve the matrix formula (9), the extreme value solution is the optimal solution and Every non-zero Lagrange multiplier α i The corresponding input vector xi is the least squares support vector falling on the classification hyperplane;

[0040] S16: Construct the optimal quasi-kernel function through the optimal solution. The formula is as follows:

[0041]

[0042] The input parameters are divided into training sets and validation sets according to the proportion using a random function.

[0043] Preferably, in S4, the kernel function parameter σ and the penalty parameter c are optimized based on the fruit fly optimization algorithm, and the calculation formula of the index is as follows:

[0044]

[0045] In the formula, MRE represents the mean relative error, MAE represents the maximum relative error, MSE represents the mean square error, R 2 represents the degree of fit, Represents the model calculated value.

[0046] Preferably, the fruit fly optimization algorithm comprises the following steps:

[0047] S21: Initialize the position of the fruit fly population P (x0, y0), the population size N and the maximum number of iterations i max parameter;

[0048] S22: Set the random direction and single flight step length L(i) of the fruit fly individual to find food. After i iterations, the position of the fruit fly individual is updated to P(x i ,y i ), where x i =x0+L(i),y i =y0+L(i);

[0049] S23: After i iterations, the distance that the fruit fly flies out is Calculate the judgment value E of taste concentration i =1 / D i ;

[0050] S24: Substitute the above smell concentration judgment value into the judgment function f(i) to calculate the smell concentration value of each fruit fly individual: Smell(i) = f(E i );

[0051] S25: Determine the taste concentration value and position of the fruit fly individual with the best taste concentration as follows:

[0052] [Smell_best, Index_best]=min(Smell) (15);

[0053] S26: Assign initial values ​​to the kernel function parameter σ and the penalty parameter c:

[0054] σ=C1E i (1,1), c = C2E i (1,2) (16);

[0055] In the formula, E i is a 1×2 vector, C1 and C2 are constants;

[0056] S27: Determine whether Smell_best(i) is less than the optimal flavor concentration Smell_best(i0). If so, keep the concentration value and the position of the fruit fly. Otherwise, record the position corresponding to Smell_best(i).

[0057] S28: Determine whether the maximum number of iterations i has been reached max , if yes, execute the next step, otherwise, execute S22-S28 in a loop;

[0058] S29: Output the optimization result kernel function parameter σ and penalty parameter c.

[0059] Preferably, different fuzzy memberships are assigned to sample points in each fuzzy subset based on the membership function, and the results of all model outputs are superimposed by the fuzzy memberships. The superimposed output is used as the final prediction result of the model, and the error value and fit between the training set data and the prediction result are calculated.

[0060] Preferably, in S5, based on the trained and verified fuzzy least squares support vector machine model, the prediction result T of the i-th temperature value in the input new sample is obtained. i , the corresponding fuzzy membership weight is μ i , the formula is as follows:

[0061]

[0062] In the formula, l represents the number of fuzzy subsets, and the output of fuzzy superposition is T I ' is the final result.

[0063] Compared with the prior art, the present invention has the following advantages:

[0064] This application introduces fuzzy methods to construct a fuzzy least squares support vector machine model, which is convenient for the effective processing of subsequent data. By optimizing the model parameters using the fruit fly optimization algorithm, the ability of the fuzzy method to process non-deterministic information is effectively brought into play. By making full use of the least squares support vector machine, the speed and generalization of the model prediction are effectively improved, making the prediction of the geothermal field more accurate, while the overall cost is lower and more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0066] Figure 1 It is the overall flow chart of the present invention;

[0067] Figure 2 It is a structural diagram of the prediction model in the present invention;

[0068] Figure 3 It is a flow chart of the fruit fly optimization algorithm in the present invention;

[0069] Figure 4 It is a schematic diagram of the correlation analysis results in the present invention;

[0070] Figure 5 It is a schematic diagram of the training set results in the present invention;

[0071] Figure 6 It is a schematic diagram of the verification set results in the present invention;

[0072] Figure 7 This is a schematic diagram of the verification structure of the fuzzy least squares support vector machine model in the present invention. Figure 1 ;

[0073] Figure 8 This is a schematic diagram of the model verification structure of the present invention Figure 2 ;

[0074] Fig. 9 This is a schematic diagram of the model verification structure of the present invention Figure 3 ;

[0075] Fig.10 This is a schematic diagram of the model verification structure of the present invention Figure 4 ;

[0076] Fig.11 It is a Gaussian function structure diagram in the present invention. DETAILED DESCRIPTION

[0077] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0078] like Figure 1-Figure 11 As shown, this embodiment proposes a method for predicting geothermal field changes based on fuzzy least squares support vector machine, including the following steps:

[0079] S1: Based on the borehole distribution map, arrange shallow geothermal monitoring points and obtain initial monitoring parameters;

[0080] S2: Based on the initial parameters, Pearson correlation analysis is performed to determine the input parameters of the prediction model;

[0081] S3: Fuzzify the input parameters and establish a fuzzy least squares support vector machine model;

[0082] S4: Optimize the parameters based on the fruit fly optimization algorithm, and perform self-training and self-validation of the fuzzy least squares support vector machine model;

[0083] S5: Predict the geothermal field based on the trained and verified fuzzy least squares support vector machine model.

[0084] In the ground source heat pump system, the prediction of ground temperature is crucial to the operation and regulation of the system. However, it is costly and difficult to obtain the deeper ground temperature. In contrast, the parameters of the machine room system and the shallow ground temperature are easy to monitor and obtain. Because there is a coupling relationship between ground temperature, system load and outdoor environmental parameters. Therefore, the deep ground temperature can be predicted by inversion.

[0085] Due to the invisibility of underground rock and soil, it is approximately a black box, but some information can be obtained through drilling exploration and rock and soil thermal response testing. Therefore, on the premise of obtaining certain information, the underground rock and soil can be treated as a gray box model and the data can be fuzzy processed to improve the accuracy of model prediction.

[0086] For building area greater than 5000m 2 According to the "Technical Specifications for Ground Source Heat Pump System Engineering 2009", large-scale ground source heat pump systems need to conduct geotechnical thermal response test experiments, and the number of test holes needs to be appropriately increased with the increase in building area. Therefore, this application adds low-cost surface ground temperature measurement points (1~5m) on the basis of conventional machine rooms and ground temperature monitoring systems, and combines the fuzzy least squares vector machine algorithm to establish a ground temperature prediction model.

[0087] In S1, combined with the borehole distribution map, shallow geothermal monitoring points are evenly arranged to determine the system operation parameter monitoring points. The initial parameters include the supply and return water temperature on the user side, the supply and return water temperature, flow rate, and ambient temperature on the ground source side. During the thermal response test, the test hole can be arranged with a complete geothermal monitoring system, that is, geothermal monitoring points are arranged within the entire borehole depth range. The machine room monitoring system provides conventional monitoring parameter data, on this basis, geothermal monitoring points are added 1 to 5 meters underground.

[0088] Sensors need to be arranged in the test hole to obtain the complete ground temperature distribution. Considering that the ground temperature distribution is distributed in variable temperature layers and normal temperature layers, a sensor is set every 1m from 1 to 10m underground, and a sensor is set every 10m from 10 to 100m. Considering the thermal hysteresis (thermal inertia) of the ground temperature response, the test cycle needs to reach a certain length, after the ground temperature has a more obvious change, to improve the sensitivity of the correlation between parameters. Take a cooling or heating season working cycle as the benchmark.

[0089] In S2, the Pearson correlation coefficient is a measure of the linear correlation between two variables, and the formula is as follows:

[0090]

[0091] In the formula, represents the average value of x, Represents the average value of y, and the calculation result r xyThe value range is between -1 and 1, where -1 indicates a completely negative correlation, 1 indicates a completely positive correlation, and 0 indicates no correlation.

[0092] Through the above analysis method, r xy Parameters whose absolute values ​​are not less than 0.5 are the input parameters of the prediction model.

[0093] Figure 4 is a schematic diagram of the correlation analysis results in the present invention, Figure 4 It can be seen that the user side / ground source side inlet and outlet water temperature and flow, ambient temperature and other parameters have a certain correlation with the ground temperature value, and the temperature of the underground surface layer 1 to 5m is also related to the ground temperature of the deeper layer 40m to 100m. The correlation with the ambient temperature and the inlet and outlet water temperature of the buried pipe is also strong.

[0094] In S3, considering the black box properties of underground rock and soil, a fuzzy algorithm is introduced to enhance the correlation between input and output parameters, and a suitable membership function is selected to fuzzify the input data, thereby converting the determined system input quantity into the corresponding fuzzy variable. This application uses a Gaussian membership function to fuzzify the input parameters, and the formula is as follows:

[0095]

[0096] In the formula, k represents the number of fuzzy sets, T represents the input sample, T0(i) and σ represent the symmetry axis and variance of the i-th membership function, respectively. Represents the fuzzy membership of the jth input data of sample i.

[0097] The method for establishing the fuzzy least squares support vector machine model includes:

[0098] S11: Define the training set as:

[0099] T={(x i ,y i )|x i ∈R m ,y i ∈R n ,i=1,2,…l} (3);

[0100] In the formula, l represents the total number of samples, R m represents the input space, R n Represents the output space;

[0101] The decision function is defined as:

[0102]

[0103] Where w∈R m represents the normal vector, b represents the threshold, Represents the mapping of input data in high-dimensional space;

[0104] S12: Transformed into solving the constrained minimum problem, the formula is as follows:

[0105]

[0106] Where J represents structural risk, y i represents the constraint condition, c represents the penalty parameter, e i represents the error variable;

[0107] S13: Introduce Lagrange multiplier α=[α1,α2,…α n ], construct the Lagrangian function, the formula is as follows:

[0108]

[0109] S14: Solve partial derivatives of w, b, e, and α respectively, and eliminate w and e i , the formula is as follows:

[0110]

[0111] Define the radial basis function (RBF) as the kernel function, the formula is as follows:

[0112]

[0113] In the formula, σ is used as the kernel function width adjustment parameter to control the range of the radial basis function (RBF);

[0114] S15: Combine equations (7) and (8) to obtain the matrix of the linear equation system, as follows:

[0115]

[0116] Solve the matrix formula (9), the extreme value solution is the optimal solution and Every non-zero Lagrange multiplier α i The corresponding input vector xi is the least squares support vector falling on the classification hyperplane;

[0117] S16: Construct the optimal quasi-kernel function through the optimal solution. The formula is as follows:

[0118]

[0119] Divide the input parameters and use a random function to divide the training set and validation set according to the proportion. The proportion of the training set should not be less than 70%.

[0120] In S4, the first is the self-training and self-validation process of the least squares support vector machine model in each fuzzy subset, that is, the model is trained and validated separately using the training set in each fuzzy subset, and the kernel function parameter σ and the penalty parameter c are optimized based on the fruit fly optimization algorithm. The calculation formula of the indicator is as follows:

[0121]

[0122]

[0123] In the formula, MRE represents the mean relative error, MAE represents the maximum relative error, MSE represents the mean square error, R 2 represents the degree of fit, Indicates the model calculation value, the superscript max represents the maximum value, the superscript min represents the minimum value of the actual value, and the superscript mean represents the average value. To ensure that the mean relative error (MRE), maximum relative error (MAE) and mean square error (MSE) are as small as possible, the goodness of fit (R 2 ) is close to 1.

[0124] The fruit fly optimization algorithm includes the following steps:

[0125] S21: Initialize the position of the fruit fly population P (x0, y0), the population size N and the maximum number of iterations i max parameter;

[0126] S22: Set the random direction and single flight step length L(i) of the fruit fly individual to find food. After i iterations, the position of the fruit fly individual is updated to P(x i ,y i ), where x i =x0+L(i),y i =y0+L(i);

[0127] S23: After i iterations, the distance that the fruit fly flies out is Calculate the judgment value E of taste concentration i =1 / D i ;

[0128] S24: Substitute the above smell concentration judgment value into the judgment function f(i) to calculate the smell concentration value of each fruit fly individual: Smell(i) = f(E i );

[0129] S25: Determine the taste concentration value and position of the fruit fly individual with the best taste concentration as follows:

[0130] [Smell_best, Index_best]=min(Smell) (15);

[0131] S26: Assign initial values ​​to the kernel function parameter σ and the penalty parameter c:

[0132] σ=C1E i (1,1), c = C2E i (1,2) (16);

[0133] In the formula, E i is a 1×2 vector, C1 and C2 are constants;

[0134] S27: Determine whether Smell_best(i) is less than the optimal flavor concentration Smell_best(i0). If so, keep the concentration value and the position of the fruit fly. Otherwise, record the position corresponding to Smell_best(i).

[0135] S28: Determine whether the maximum number of iterations i has been reached max , if yes, execute the next step, otherwise, execute S22-S28 in a loop;

[0136] S29: Output the optimization result kernel function parameter σ and penalty parameter c.

[0137] The fruit fly optimization algorithm is used to optimize the kernel function parameter σ and the penalty parameter c, including setting the population number, population size and maximum number of iterations, and then the fitting degree R of the training data and the verification data is calculated. 2 As the taste concentration judgment function, that is, f(E i )=R 2 At the end of the iteration, f(E i ) The optimal concentration corresponding to the minimum value E i , which is the global optimal solution of σ and c.

[0138] Then comes the self-training and self-validation process of the model in all training data sets, that is, different fuzzy memberships are assigned to the sample points in each fuzzy subset based on the membership function, and the results of all model outputs are superimposed through the fuzzy membership. The superimposed output is used as the final prediction result of the model, and the error value and fit between the training set data and the prediction result are calculated.

[0139] The last step is the generalization verification process of the model in the validation data set, that is, the model input is based on the validation set data, and compared with the actual value, and the error and fit between the predicted value and the actual value are calculated. The accuracy of the verification result directly reflects the quality of the model prediction effect.

[0140] In S5, based on the trained and validated fuzzy least squares support vector machine model, the predicted result T of the i-th temperature value of the input new sample is obtained. i , the corresponding fuzzy membership weight is μ i , the formula is as follows:

[0141]

[0142] In the formula, l represents the number of fuzzy subsets, and the output of fuzzy superposition is T I ' is the final result.

[0143] Due to the large amount of system monitoring data, it is difficult to directly use the least squares support vector machine to process the data. Therefore, by introducing the fuzzy method, a fuzzy-based least squares support vector machine model is constructed to facilitate the effective processing of subsequent data. By optimizing the model parameters using the fruit fly optimization algorithm, the ability of the fuzzy method to process uncertain information is effectively brought into play. By making full use of the least squares support vector machine, the rapidity and generalization of the model prediction are effectively improved, making the prediction of the geothermal field more accurate, while the overall cost is lower and more reliable.

[0144] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for predicting geothermal field changes based on fuzzy least squares support vector machine, characterized in that: The following steps are involved: S1: Based on the borehole distribution map, arrange shallow geothermal monitoring points and obtain initial monitoring parameters; S2: Based on the initial parameters, Pearson correlation analysis is performed to determine the input parameters of the prediction model; S3: Fuzzify the input parameters and establish a fuzzy least squares support vector machine model; S4: Optimize the parameters based on the fruit fly optimization algorithm, and perform self-training and self-validation of the fuzzy least squares support vector machine model; S5: Predict the geothermal field based on the trained and verified fuzzy least squares support vector machine model.

2. The method for predicting ground temperature field changes based on fuzzy least squares support vector machine according to claim 1 is characterized in that: In S1, the initial parameters include the supply and return water temperature on the user side, the supply and return water temperature on the ground source side, the flow rate, and the ambient temperature.

3. The method for predicting ground temperature field changes based on fuzzy least squares support vector machine according to claim 2 is characterized in that: In S2, the Pearson correlation coefficient is given by the following formula: In the formula, represents the average value of x, represents the mean value of y; Determine r xy Parameters whose absolute values ​​are not less than 0.5 are the input parameters of the prediction model.

4. The method for predicting ground temperature field changes based on fuzzy least squares support vector machine according to claim 1 is characterized in that: In S3, the Gaussian membership function is used to perform fuzzy processing on the input parameters, and the formula is as follows: In the formula, k represents the number of fuzzy sets, T represents the input sample, T0(i) and σ represent the symmetry axis and variance of the i-th membership function, respectively. Represents the fuzzy membership of the jth input data of sample i.

5. The method for predicting ground temperature field changes based on fuzzy least squares support vector machine according to claim 4 is characterized in that: The method for establishing the fuzzy least squares support vector machine model includes: S11: Define the training set as: T={(x i ,y i )|x i ∈R m ,y i ∈R n ,i=1,2,…l} (3); In the formula, l represents the total number of samples, R m represents the input space, R n Represents the output space; The decision function is defined as: Where w∈R m represents the normal vector, b represents the threshold, Represents the mapping of input data in high-dimensional space; S12: Transformed into solving the constrained minimum problem, the formula is as follows: Where J represents structural risk, y i represents the constraint condition, c represents the penalty parameter, e i represents the error variable; S13: Introduce Lagrange multiplier α=[α1,α2,…α n ], construct the Lagrangian function, the formula is as follows: S14: Solve partial derivatives of w, b, e, and α respectively, and eliminate w and e i , the formula is as follows: Define the radial basis function as the kernel function, the formula is as follows: In the formula, σ is used as the kernel function width adjustment parameter to control the range of the radial basis function; S15: Combine equations (7) and (8) to obtain the matrix of the linear equation system, as follows: Solve the matrix formula (9), the extreme value solution is the optimal solution and Every non-zero Lagrange multiplier α i The corresponding input vector xi is the least squares support vector falling on the classification hyperplane; S16: Construct the optimal quasi-kernel function through the optimal solution. The formula is as follows: The input parameters are divided into training sets and validation sets according to the proportion using a random function.

6. The method for predicting ground temperature field changes based on fuzzy least squares support vector machine according to claim 1 is characterized in that: In S4, the kernel function parameter σ and the penalty parameter c are optimized based on the fruit fly optimization algorithm, and the calculation formula of the index is as follows: In the formula, MRE represents the mean relative error, MAE represents the maximum relative error, MSE represents the mean square error, R 2 represents the degree of fit, Represents the model calculated value.

7. The method for predicting ground temperature field changes based on fuzzy least squares support vector machine according to claim 6 is characterized in that: The fruit fly optimization algorithm includes the following steps: S21: Initialize the position of the fruit fly population P (x0, y0), the population size N and the maximum number of iterations i max parameter; S22: Set the random direction and single flight step length L(i) of the fruit fly individual to find food. After i iterations, the position of the fruit fly individual is updated to P(x i ,y i ), where x i =x0+L(i),y i =y0+L(i); S23: After i iterations, the distance that the fruit fly flies out is Calculate the judgment value E of taste concentration i =1 / D i ; S24: Substitute the above smell concentration judgment value into the judgment function f(i) to calculate the smell concentration value of each fruit fly individual: Smell(i) = f(E i ); S25: Determine the taste concentration value and position of the fruit fly individual with the best taste concentration as follows: [Smell_best, Index_best]=min(Smell) (15); S26: Assign initial values ​​to the kernel function parameter σ and the penalty parameter c: σ=C1E i (1,1), c=C2E i (1,2) (16); In the formula, E i is a 1×2 vector, C1 and C2 are constants; S27: Determine whether Smell_best(i) is less than the optimal flavor concentration Smell_best(i0). If so, keep the concentration value and the position of the fruit fly. Otherwise, record the position corresponding to Smell_best(i). S28: Determine whether the maximum number of iterations i has been reached max , if yes, execute the next step, otherwise, execute S22-S28 in a loop; S29: Output the optimization result kernel function parameter σ and penalty parameter c.

8. The method for predicting ground temperature field changes based on fuzzy least squares support vector machine according to claim 7 is characterized in that: Based on the membership function, different fuzzy memberships are assigned to the sample points in each fuzzy subset. The results of all model outputs are superimposed through the fuzzy memberships. The superimposed output is used as the final prediction result of the model, and the error value and fit between the training set data and the prediction result are calculated.

9. The method for predicting ground temperature field changes based on fuzzy least squares support vector machine according to claim 1, characterized in that: In S5, based on the trained and verified fuzzy least squares support vector machine model, the prediction result T of the i-th temperature value of the input new sample is obtained. i , the corresponding fuzzy membership weight is μ i , the formula is as follows: In the formula, l represents the number of fuzzy subsets, and the output of fuzzy superposition is T I ' is the final result.

Images