Method for manufacturing and optimizing comprehensive pointing model of horizon telescope
The comprehensive pointing model for horizontal telescopes, integrating basic parameter and spherical harmonic terms with polynomial regression and machine learning, enhances pointing accuracy for high-precision astronomical observations.
Patent Information
- Application Number
- CN202411200346.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-12-06
AI Technical Summary
Existing pointing models for horizontal telescopes, such as basic parameter models and spherical harmonic function models, struggle to accurately correct all influencing factors, limiting high-precision astronomical observation accuracy.
A comprehensive pointing model is developed using a combination of basic parameter terms, spherical harmonic terms, polynomial regression, and machine learning to dynamically adjust parameters for real-time correction of telescope pointing deviations.
Significantly improves the observation accuracy of horizontal telescopes by effectively correcting pointing errors, meeting the demands of high-precision astronomical observations.
Smart Images

Figure CN119089307A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of astronomical observation equipment, and in particular to a method for optimizing and correcting a pointing model of an astronomical telescope. The method is applied to astronomical telescopes, pointing and positioning turntables and related fields to improve the pointing accuracy of astronomical telescopes. Background Art
[0002] With the advancement of science and technology, astronomical observation technology has developed rapidly. The observation capabilities of modern astronomical telescopes are constantly improving, and the accuracy requirements for observation data are getting higher and higher. In order to meet the needs of high-precision observation, the pointing accuracy of the telescope must be strictly controlled and corrected. The horizontal telescope has become the main equipment for astronomical observation because of its relatively simple structure and convenient operation. Its pointing accuracy is a key indicator to measure the accuracy of the telescope's observation position of the target. At present, there are methods to correct pointing errors using basic parameter models and spherical harmonic function models respectively. The correction effects of these two pointing models have their own advantages. The correction ability of the basic parameter model is limited, and it is difficult to fully consider all influencing factors. It can only fit the error terms with clear physical meanings generated by the telescope. In contrast, the spherical harmonic function model performs well in dealing with high-order error terms, and it can accurately fit complex errors. Summary of the invention
[0003] In view of the above problems existing in the prior art, the present invention provides a method for manufacturing and optimizing a comprehensive pointing model of a horizontal telescope.
[0004] To achieve the above object, the present invention provides the following technical solutions:
[0005] The method for making and optimizing the comprehensive pointing model of a horizontal telescope comprises the following steps:
[0006] Step 1: Use the basic parameter terms of the basic parameter model and the spherical harmonic terms of the spherical harmonic function model to form an error correction model;
[0007] Step 2: Use polynomial regression and machine learning methods to create a polynomial regression model;
[0008] Step 3: Use a comprehensive pointing model consisting of an error correction model and a polynomial regression model to correct the telescope pointing deviation.
[0009] Furthermore, before step 1, there are also data acquisition and preprocessing steps and error analysis and screening steps;
[0010] The data acquisition and preprocessing steps include: inputting pointing instructions to the telescope according to a predetermined azimuth interval and elevation interval, and the azimuth and elevation of the telescope are saved as actual values; taking multiple sets of star images, and obtaining the right ascension and declination corresponding to the centers of the multiple sets of star images by comparing with the standard star images; converting the right ascension and declination into the azimuth and elevation in the horizontal coordinate system, i.e., the theoretical values of the azimuth and elevation, according to the recorded data in the star images; and calculating the deviation between the actual value and the theoretical value;
[0011] The error analysis and screening step includes: analyzing the deviation data obtained in the data acquisition and preprocessing steps, and screening out error points that exceed a set threshold.
[0012] Furthermore, in step 1, the basic parameter model is established based on physical factors, the model is stable and each parameter has a clear physical meaning; the spherical harmonic function model is used to process high-order error terms; the combined error correction model is used to dynamically adjust parameters of real-time observation data to adapt to different observation conditions and equipment states.
[0013] Furthermore, in step 1, the azimuth pointing error ΔA and the elevation pointing error ΔH are expressed as:
[0014]
[0015] ΔH=IE-AN·cosA+AW·sinA-TX·tan(90°-Z)
[0016] +e0+e1·sinA+e2·cosA+e3·sinH+e4·cosH
[0017] +e5·cosA·sinH+e6·cosA·sin 2 H+e7·cos 3 A
[0018] Among them, A, H, and Z represent the actual azimuth, elevation, and zenith distance of the telescope, respectively; IA, IE, NPAE, CA, AN, AW, and TX are the coefficients of the basic parameter items; IA is the zero-point error of the azimuth, IE is the zero-point error of the elevation, NPAE represents the non-orthogonality between the azimuth axis and the elevation axis, CA represents the non-orthogonality between the barrel axis and the elevation axis, AN represents the north-south tilt of the azimuth axis, AW represents the east-west tilt of the azimuth axis, and TX represents the bending of the barrel; q0, q1, ..., q 12 are the coefficients of the azimuth spherical harmonics, and e0, e1, ..., e7 are the coefficients of the elevation spherical harmonics. After the coefficients are fitted based on the actual errors, for each set of observations, the telescope control system calculates the possible errors based on this model and automatically corrects them.
[0019] Furthermore, after the form of the error correction model is determined in step 1, the coefficient of each item is solved, and the coefficient of each item is obtained by fitting using the actual error and the actual azimuth and elevation angles through the least squares method, and then the error correction model is obtained. By adjusting the coefficients of the error correction model, the sum of the squares of the differences between the model prediction value and the actual value is minimized.
[0020] Furthermore, in step 2, the residual residual is fitted using a polynomial regression model with automatic order determination, and the mean square error RMSE and R are calculated after each fitting. 2 , use the two as scores to judge, save the model with the highest score, and calculate the theoretical residual based on this model.
[0021] Furthermore, the specific steps for automatically determining the order are as follows:
[0022] Step 2-1: Import necessary libraries: (1) Import the numpy library for numerical calculations; (2) Import PolynomialFeatures, StandardScaler, LinearRegression, train_test_split, mean_squared_error, and r2_score from the sklearn library for polynomial regression, data standardization, linear regression, data segmentation, mean squared error calculation, and R 2 Calculation; (3) Import the joblib library for saving and loading models;
[0023] Step 2-2: Data preprocessing and segmentation: (1) Create a function polynomial_regression, which accepts the feature matrix X consisting of the actual azimuth angle sj_AZ and the corresponding actual pitch angle sj_ALT, the residual error residual_error_ALT that still exists after the error correction model is fitted as the target variable y, and sets the maximum polynomial degree max_degree; (2) Use StandardScaler to standardize the training set and test set and scale the data to the same range; (3) Use the train_test_split function to split the data set into a training set and a test set;
[0024] Step 2-3: Polynomial regression model training and optimization: (1) Initialize the best score, best number of times, best model, and best R 2 Examples of fractional and optimal polynomial feature conversion; (2) Use joblib to save the optimal polynomial regression model, normalizer, and polynomial features to a local file for later loading and use;
[0025] Step 2-4: Call the model to predict residuals: (1) Use joblib to load the saved model, normalizer, and polynomial feature transformation instance; (2) Normalize the new data and perform polynomial feature transformation; (3) Use the loaded model to make predictions and return the prediction results.
[0026] Furthermore, in step 2-3, for each polynomial degree from 1 to max_degree, perform the following operations in sequence:
[0027] Step 2-3-1: Use PolynomialFeatures to perform polynomial feature conversion;
[0028] Step 2-3-2: Use LinearRegression to perform linear regression fitting;
[0029] Step 2-3-3: Calculate the mean square error and R on the test set 2 Fraction;
[0030] Step 2-3-4: If the current mean square error is less than the best score, update the best score, best number of times, best model, and best R 2 fractions, and the corresponding polynomial features.
[0031] Furthermore, step 3 also includes an overall evaluation and correction step, which calculates the residual between the fitting result and the actual error, draws a comparison chart between the actual error and the fitting result, and intuitively displays the fitting effect of the pointing model; by evaluating the goodness of fit, the accuracy and reliability of the model are comprehensively judged.
[0032] Compared with the prior art, the present invention has the following beneficial effects:
[0033] The present invention provides an optimization method. First, the basic parameter terms and spherical harmonic terms are used to form an error correction model, then the polynomial regression model is made by using polynomial regression and machine learning technology, and finally the comprehensive pointing model composed of the error correction model and the polynomial regression model is used to correct the telescope pointing deviation. After inspection, the method of the present invention can significantly improve the observation accuracy of the horizontal telescope and meet the needs of high-precision astronomical observation. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 is a flow chart of the method of the present invention;
[0035] Figure 2 is a method flow chart of Step 1 in the embodiment;
[0036] Figure 3 It is a flow chart of the method for automatically determining the order in polynomial regression model training;
[0037] Figure 4 It is a comparison chart of the actual error of AZ and the fitting result;
[0038] Figure 5 This is a comparison chart of the actual error of ALT and the fitting result. DETAILED DESCRIPTION
[0039] The present invention will be further described in detail below in conjunction with the accompanying drawings.
[0040] like Figure 1 As shown, the present invention realizes the production and optimization of the pointing model of the horizontal telescope through the following steps:
[0041] Step 1. Data acquisition and preprocessing (such as Figure 2 (shown)
[0042] First, input pointing instructions to the telescope at intervals of about 30° in azimuth and about 10° in elevation. At this time, the azimuth and elevation of the telescope are saved as actual values. A total of 76 sets of star images were taken, and by comparing with the standard star map, the right ascension and declination corresponding to the center of the 76 sets of images were obtained. Then, according to the shooting time recorded in the image, the geographical longitude and latitude of the observatory site, the altitude, air pressure and temperature, the right ascension and declination were converted into azimuth and elevation in the horizontal coordinate system, that is, the theoretical azimuth and theoretical elevation. Calculate the deviation between the actual value and the theoretical value and save it in a spreadsheet for subsequent calculation and analysis.
[0043] Step 2. Error analysis and screening
[0044] Analyze the calculated deviation data and filter out the error points that exceed the set threshold. Setting a reasonable error threshold helps to identify observation data points with significant errors, which may be caused by mechanical errors, environmental influences, etc. These errors can be further analyzed to improve the accuracy of the model.
[0045] Step 3. Build an error correction model
[0046] The error correction model is a model that uses the actual azimuth and elevation angles to predict errors, and performs advance corrections on the telescope pointing. After analyzing the telescope structure and the main error sources, the basic parameter terms and spherical harmonic terms are obtained. A, H, and Z represent the actual azimuth, elevation, and zenith distance of the telescope, respectively. The azimuth pointing error ΔA and the elevation pointing error ΔH can be expressed as:
[0047]
[0048] ΔH=IE-AN·cosA+AW·sinA-TX·tan(90°-Z)
[0049] +e0+e1·sinA+e2·cosA+e3·sinH+e4·cosH
[0050] +e5·cosA·sinH+e6·cosA·sin 2 H+e7·cos 3 A
[0051] Among them, IA, IE, NPAE, CA, AN, AW, TX are the coefficients of the basic parameter items, IA is the zero-point error of the azimuth angle, IE is the zero-point error of the elevation angle, NPAE represents the non-orthogonality between the azimuth axis and the elevation axis, CA represents the non-orthogonality between the barrel axis and the elevation axis, AN represents the north-south tilt of the azimuth axis, AW represents the east-west tilt of the azimuth axis, and TX represents the barrel bending. q0, q1, ..., q 12 are the coefficients of the azimuth spherical harmonics, and e0, e1, ..., e7 are the coefficients of the elevation spherical harmonics. After the coefficients are fitted based on the actual errors, for each set of observations, the telescope control system can calculate the possible errors based on this model and automatically correct them.
[0052] In Python, it is constructed as a matrix for operation. Assuming there are m groups of data and n coefficients, an m*n matrix can be constructed. Take the first row of the error correction matrix of azimuth AZ as an example: A[0], H[0] means that the observations of the first group are substituted, a (1 ,1),...a (1,5) is the basic parameter item of the first set of observation data, a (1 ,6),...a (1,18) is the spherical harmonic term of the first set of observation data, where a (1,j) Indicates that the element is located in the 1st row and the jth column of the matrix, for example, a (1,1) Represents the element at row 1 and column 1 of the matrix.
[0053]
[0054] a (1,4) =-AN·sin(A[0])·tan(H[0]); a (1,5) =-AW·cos(A[0])·tan(H[0]);
[0055]
[0056] By analogy, we can obtain the basic parameter terms and spherical harmonic terms of the 2nd to 76th groups of observation data. There are 18 basic parameter term coefficients and spherical harmonic term coefficients in total, which are organized into a 76*18 matrix.
[0057]
[0058] Each row of the matrix corresponds to a set of error correction formulas for observed data. Adding the first row together, we get ΔA = a (1,1) +a (1,2) +...+a (1,18) , the possible azimuth pointing deviation of the first set of observation data can be predicted.
[0059] The present invention adopts a combination of a basic parameter model and a spherical harmonic function model. The basic parameter model is established based on physical factors such as equipment axis system error and encoder error. The model is stable and each parameter has a clear physical meaning. However, the correction accuracy of this model is often limited, and it is difficult to fully consider all factors that affect the pointing accuracy. Therefore, a spherical harmonic function model is used to deal with high-order error terms. While ensuring that the physical meaning of the model is clear, the flexibility of spherical harmonic functions can be used to more accurately fit complex errors. This combination of the two methods can dynamically adjust parameters according to real-time observation data, adapt to different observation conditions and equipment states, and has wide applicability.
[0060] Step 4. Error correction model coefficient fitting
[0061] After determining the form of the error correction model, it is necessary to solve the coefficients of each term -IA, -NPAE, -CA, -AN, -AW, q0, q1, ..., q 11 ,q 12 . Using the actual error and the actual azimuth and elevation angles, the coefficients of each item can be obtained by fitting through the least squares method, and then the error correction model can be obtained. The least squares method is a mathematical statistical method, the core of which is to minimize the sum of squares of errors in order to match the best function of the data. In this process, the coefficients of the error correction model are adjusted to minimize the sum of squares of the differences between the model prediction value and the actual value.
[0062] Step 5. Polynomial regression model training
[0063] After the error correction model has been used to perform a preliminary fit on the data, in order to further improve the pointing accuracy, the present invention uses a polynomial regression model to fit the remaining residuals again. Polynomial regression is an extension of linear regression. Linear regression uses a straight line to fit data, such as a linear function y=kx+b, etc., while polynomial regression uses a curve to fit data, such as a quadratic function y=ax 2 +bx+c, cubic function y=ax 3 +bx 2+cx+d and multiple functions are used to fit the data. The input of the polynomial regression model is a feature matrix consisting of the actual azimuth angle and the corresponding actual pitch angle, and the output of the model is the residual that still exists after the error correction model is fitted. These residuals are the targets that the polynomial regression model needs to continue fitting. In order to find the most suitable model, try to use 1st to 15th order polynomial regression for fitting. After each fitting, calculate the mean square error (MSE) and R 2 , use the two as scores to judge and save the model with the highest score, and calculate the theoretical residual based on this model.
[0064] The present invention uses two performance indicators, goodness of fit and comprehensive RMS value, for verification. Goodness of fit is one of the important indicators for evaluating model performance, which reflects the degree of match between the model prediction value and the actual observation value. It has many different calculation methods, the most common of which is to use the determination coefficient R 2 , also called the goodness-of-fit coefficient. R 2 The calculation formula is:
[0065]
[0066] Among them, y i is the actual value of the i-th group of observations, is the predicted value of the i-th group of observations, is the average of all observations, and n is the number of groups of observations. 2 The value of is between 0 and 1. The closer the value is to 1, the better the model fits the data.
[0067] However, when the goodness of fit is very high, the generalization ability of the model will deteriorate, and for data under nonlinear relationships, the traditional goodness of fit index cannot accurately reflect the fitting effect of the model. Therefore, on this basis, the root mean square error (RMSE) is added to reflect the accuracy of the model prediction by calculating the mean of the square of the difference between the predicted value and the true value. The smaller the RMSE value, the higher the model fitting accuracy. Assuming that we have n groups of observation data, each group of observation data has a corresponding prediction value, then the calculation formula of RMSE is:
[0068]
[0069] Among them, Y pred,i represents the predicted value of the i-th observation, Y obs,irepresents the actual value of the ith observation, and n represents the number of observations. It is not difficult to see from the calculation formula of the RMSE value that the calculation involves the square of the error, so it is very sensitive to large prediction errors. This means that if the model's predicted values at certain observation points deviate significantly from the actual values, these large errors will be magnified in the RMSE value, and in the research process, we can directly pay attention to and improve these points with poor prediction results.
[0070] The present invention uses a polynomial regression model that automatically determines the appropriate order to fit the residual. The residual plot analysis after the error correction model is fitted finds that the residual has a nonlinear relationship, but the order is unknown. Therefore, the present invention uses a polynomial regression model that automatically determines the appropriate order to fit the error data, and uses polynomials from 1st to 15th order for fitting, and calculates the mean square error MSE and the determination coefficient R 2 , and save the model with the highest score as the score for comparison. Figure 3 As shown, the specific steps to achieve automatic judgment of the order are as follows:
[0071] Step 5-1. Import necessary libraries
[0072] (1) Import the numpy library for numerical calculations.
[0073] (2) Import PolynomialFeatures, StandardScaler, LinearRegression, train_test_split, mean_squared_error, and r2_score from the sklearn library for polynomial regression, data standardization, linear regression, data segmentation, mean squared error calculation, and R 2 calculate.
[0074] (3) Import the joblib library to save and load models.
[0075] Step 5-2. Data preprocessing and segmentation
[0076] (1) Create a function polynomial_regression, which accepts the feature matrix X consisting of the actual azimuth angle sj_AZ and the corresponding actual pitch angle sj_ALT, the residual error residual_error_ALT that still exists after the error correction model is fitted as the target variable y, and sets the maximum polynomial degree max_degree to 5.
[0077] (2) Use StandardScaler to standardize the training set and test set and scale the data to the same range.
[0078] (3) Use the train_test_split function to split the dataset into a training set and a test set, where the test set accounts for 10%.
[0079] Step 5-3. Polynomial regression model training and optimization
[0080] (1) Initialize the best score, best number of times, best model, and best R 2 Examples of fractional and optimal polynomial feature transformations.
[0081] (2) For each polynomial degree from 1 to max_degree, perform the following operations in sequence:
[0082] ①Use PolynomialFeatures to perform polynomial feature conversion.
[0083] ②Use LinearRegression for linear regression fitting.
[0084] ③Calculate the mean square error and R on the test set 2 Fraction.
[0085] ④If the current mean square error is less than the best score, update the best score, best number of times, best model, and best R 2 fractions, and the corresponding polynomial features.
[0086] (3) Use joblib to save the best polynomial regression model, normalizer, and polynomial features to a local file for easy loading and use later.
[0087] Step 5-4. Call the model to predict the residual
[0088] (1) Use joblib to load the saved model, normalizer, and polynomial feature transformation instances.
[0089] (3) Standardize the new data and perform polynomial feature transformation.
[0090] (3) Use the loaded model to make predictions and return the prediction results.
[0091] Step 6. Overall evaluation and correction
[0092] By calculating the residual between the fitting result and the actual error, a comparison chart of the actual error and the fitting result is drawn, such as Figure 4-5 As shown in the figure, the fitting effect of the pointing model is intuitively displayed. The accuracy and reliability of the model are judged by evaluating the goodness of fit and the comprehensive RMSE value.
[0093] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. Preparation and optimization method of comprehensive pointing model for horizontal telescope, It is characterized in that The steps include: Step 1: Use the basic parameter terms of the basic parameter model and the spherical harmonic terms of the spherical harmonic function model to form an error correction model; Step 2: Use polynomial regression and machine learning methods to create a polynomial regression model; Step 3: Use a comprehensive pointing model consisting of an error correction model and a polynomial regression model to correct the telescope pointing deviation.
2. The method for making and optimizing the comprehensive pointing model of a horizontal telescope according to claim 1, It is characterized in that Before step 1, there are also data acquisition and preprocessing steps as well as error analysis and screening steps; The data acquisition and preprocessing steps include: inputting pointing instructions to the telescope according to a predetermined azimuth interval and elevation interval, and the azimuth and elevation of the telescope are saved as actual values; taking multiple sets of star images, and obtaining the right ascension and declination corresponding to the centers of the multiple sets of star images by comparing with the standard star images; converting the right ascension and declination into the azimuth and elevation in the horizontal coordinate system, i.e., the theoretical values of the azimuth and elevation, according to the recorded data in the star images; and calculating the deviation between the actual value and the theoretical value; The error analysis and screening step includes: analyzing the deviation data obtained in the data acquisition and preprocessing steps, and screening out error points that exceed a set threshold.
3. The method for making and optimizing the comprehensive pointing model of a horizontal telescope according to claim 1, It is characterized in that In step 1, the basic parameter model is established based on physical factors, the model is stable and each parameter has a clear physical meaning; the spherical harmonic function model is used to process high-order error terms; the combined error correction model is used to dynamically adjust parameters of real-time observation data to adapt to different observation conditions and equipment states.
4. The method for making and optimizing the comprehensive pointing model of a horizontal telescope according to claim 1 or 3, It is characterized in that In step 1, the azimuth pointing error ΔA and the elevation pointing error ΔH are expressed as: ΔH=IE-AN·cosA+AW·sinA-TX·tan(90°-Z) +e 0 +e 1 ·sinA+e 2 ·cosA+e 3 ·sinH+e 4 ·cosH +e 5 ·cosA·sinH+e 6 ·whatA·sin 2 H+e 7 ·what 3 TO Among them, A, H, and Z represent the actual azimuth, elevation, and zenith distance of the telescope, respectively; IA, IE, NPAE, CA, AN, AW, and TX are coefficients of basic parameter items; IA is the zero-point error of the azimuth, IE is the zero-point error of the elevation, NPAE represents the non-orthogonality between the azimuth axis and the elevation axis, CA represents the non-orthogonality between the barrel axis and the elevation axis, AN represents the north-south tilt of the azimuth axis, AW represents the east-west tilt of the azimuth axis, and TX represents the bending of the barrel; q 0 ,q 1 ,...,q 12 is the coefficient of the azimuthal spherical harmonic, e 0 ,e 1 ,...,e 7 are the coefficients of the spherical harmonics of the pitch angle; after the coefficients are fitted based on the actual errors, for each set of observations, the telescope control system calculates the possible errors based on this model and automatically corrects them.
5. The method for making and optimizing the comprehensive pointing model of a horizontal telescope according to claim 1, It is characterized in that After the form of the error correction model is determined in step 1, the coefficient of each item is solved. The actual error and the actual azimuth and elevation angles are used to fit the coefficient of each item through the least squares method, and then the error correction model is obtained. By adjusting the coefficients of the error correction model, the sum of the squares of the differences between the model prediction value and the actual value is minimized.
6. The method for making and optimizing the comprehensive pointing model of a horizontal telescope according to claim 1, It is characterized in that In step 2, the residual residual is fitted using a polynomial regression model with automatic order determination, and the mean square error RMSE and R are calculated after each fitting. 2 , use the two as scores to judge, save the model with the highest score, and calculate the theoretical residual based on this model.
7. The method for making and optimizing the comprehensive pointing model of a horizontal telescope according to claim 6, It is characterized in that The specific steps for automatically determining the order are as follows: Step 2-1: Import necessary libraries: (1) Import the numpy library for numerical calculations; (2) Import PolynomialFeatures, StandardScaler, LinearRegression, train_test_split, mean_squared_error, and r2_score from the sklearn library for polynomial regression, data standardization, linear regression, data segmentation, mean squared error calculation, and R 2 Calculation; (3) Import the joblib library for saving and loading models; Step 2-2: Data preprocessing and segmentation: (1) Create a function polynomial_regression, which accepts the feature matrix X consisting of the actual azimuth angle sj_AZ and the corresponding actual pitch angle sj_ALT, the residual error residual_error_ALT that still exists after the error correction model is fitted as the target variable y, and sets the maximum polynomial degree max_degree; (2) Use StandardScaler to standardize the training set and test set and scale the data to the same range; (3) Use the train_test_split function to split the dataset into training set and test set; Step 2-3: Polynomial regression model training and optimization: (1) Initialize the best score, best number of times, best model, and best R 2 Examples of fractional and optimal polynomial feature conversion; (2) Use joblib to save the optimal polynomial regression model, normalizer, and polynomial features to a local file for later loading and use; Step 2-4: Call the model to predict residuals: (1) Use joblib to load the saved model, normalizer, and polynomial feature transformation instance; (2) Normalize the new data and perform polynomial feature transformation; (3) Use the loaded model to make predictions and return the prediction results.
8. The method for making and optimizing the comprehensive pointing model of a horizontal telescope according to claim 7, It is characterized in that In step 2-3, for each polynomial degree from 1 to max_degree, perform the following operations in sequence: Step 2-3-1: Use PolynomialFeatures to perform polynomial feature conversion; Step 2-3-2: Use LinearRegression to perform linear regression fitting; Step 2-3-3: Calculate the mean square error and R on the test set 2 Fraction; Step 2-3-4: If the current mean square error is less than the best score, update the best score, best number of times, best model, and best R 2 fractions, and the corresponding polynomial features.
9. The method for making and optimizing the comprehensive pointing model of a horizontal telescope according to claim 1, It is characterized in that After step 3, an overall evaluation and correction step is also included. By calculating the residual between the fitting result and the actual error, a comparison chart of the actual error and the fitting result is drawn to intuitively display the fitting effect of the pointing model; by evaluating the goodness of fit, the accuracy and reliability of the model are comprehensively judged.
