CLSM fluorescence spectrum-based detection and visualization method for density of crude oil in rock slice

Abstract

The invention discloses a method for detecting and visualizing the density of crude oil in a rock slice based on a laser scanning confocal microscope (CLSM) fluorescence spectrum, and relates to the field of crude oil density detection. The method comprises the following steps: collecting crude oil spectral data by using CLSM, and carrying out baseline correction, data smoothing and standardization processing; spectral features highly related to the density of the crude oil are screened out through the Pearson's correlation coefficient; a density prediction model is established based on an SVM-R model, and parameters are optimized to improve prediction precision. The density prediction and result visualization of the spectral data are realized by utilizing self-developed interactive software. According to the method, the detection precision R2 of a complex sample is remarkably improved to 0.9558, and the method has real-time analysis and dynamic visualization functions and is suitable for crude oil density evaluation and thermal drive optimization scenes in oilfield development.

Description

Technical Field

[0001] The present invention relates to the technical field of crude oil density detection, and in particular to a method for density prediction and visualization using a laser scanning confocal microscope to collect spectral data and combining a support vector machine regression (SVM) model. Background Art

[0002] Crude oil density is an important indicator of crude oil quality and has a significant impact on market pricing, refining processes and the formulation of oilfield development plans. Traditional crude oil density measurement methods, such as the standard density meter method (SH / T0604-2000) and the U-shaped vibrating tube method, have high measurement accuracy but are complex and time-consuming to operate and rely on strict laboratory conditions, making it difficult to meet the needs of rapid testing and on-site evaluation.

[0003] In recent years, spectral analysis technology has been widely used in crude oil density prediction due to its advantages of fast testing speed, non-destructive detection and suitability for complex sample analysis. Among them, transmission infrared spectroscopy technology extracts specific bands (750-1350cm -1 ) absorbance data, combined with chemometric models to predict crude oil density (such as patent CN 103063599A). However, infrared spectroscopy is sensitive to sample matrix effects, especially in samples with high viscosity or complex components, the prediction accuracy is significantly reduced, and it is impossible to visualize density changes in real time, which limits its application in reservoir dynamic monitoring.

[0004] Fluorescence spectroscopy has gradually attracted attention in recent years. It can more comprehensively reflect the physical properties of crude oil by analyzing the excitation and emission characteristics of fluorescent active molecules in the sample. Some studies have applied fluorescence spectroscopy to crude oil density prediction (such as patent CN 118329692 A), using the QGF-E wavelength of the fluorescence spectrum of reservoir particle extracts to establish a regression model to predict crude oil density. However, this method relies on reservoir particle extraction operations, the experimental process is relatively complicated, and the model applicability is highly dependent on the region. In addition, the technology lacks real-time visualization capabilities and is difficult to meet the needs of dynamic density assessment in complex scenarios.

[0005] Laser confocal microscopy technology has also been used in micro-area spectral detection due to its high spatial resolution and sensitivity (such as patent CN 107167400 A). This technology predicts sample density by fitting spectral data, but it is mainly aimed at specific scenarios, the model has a narrow scope of application, and the prediction accuracy (correlation coefficient is about 0.91) is not enough to meet high-precision requirements. At the same time, this technology has limitations in result display and real-time performance, and cannot achieve real-time analysis and visualization of crude oil density in dynamic development processes such as thermal drive.

[0006] In summary, the existing technologies have obvious deficiencies in the following aspects: 1. Applicability limitations: some technologies are applicable to specific scenarios or sample types and lack a general model with strong universality; 2. Insufficient prediction accuracy: the prediction accuracy of existing methods in complex samples (such as high-viscosity crude oil and heavy oil) still has room for improvement; 3. Insufficient dynamic evaluation capabilities: the existing technologies cannot meet the real-time analysis needs of crude oil density changes during dynamic reservoir development; 4. The results are not intuitive: the lack of interactive visualization tools limits the practical application and efficiency of density prediction results. Summary of the invention

[0007] This application aims to solve the shortcomings of traditional crude oil density detection methods, such as long time consumption and reliance on complex laboratory operations. For example, infrared spectroscopy technology has poor adaptability to high viscosity and complex samples, and is easily interfered by matrix effects, thus affecting the prediction accuracy; and although laser scanning confocal technology is suitable for micro-area detection, its crude oil density prediction accuracy is low (the correlation coefficient is about 0.91), and the above methods cannot achieve real-time visualization of crude oil density in rock slices during thermal drive, and it is difficult to clarify the optimal temperature conditions for thermal drive in oil field development. The present application provides a method for detecting and visualizing crude oil density in rock slices based on CLSM fluorescence spectroscopy. The method should have both ease of operation, high-precision prediction capabilities and a wide range of application scenarios to meet the needs of modern petroleum industry for rapid evaluation, dynamic monitoring and optimized decision-making; a crude oil density detection method that is efficient, accurate, highly applicable and has real-time visualization function.

[0008] The present invention adopts the following technical solutions to solve the above technical problems:

[0009] The technical solution adopted by the present invention is as follows: a method for detecting and visualizing the density of crude oil in rock slices based on CLSM fluorescence spectroscopy, comprising the following steps:

[0010] (1) Data collection: spectral data of crude oil samples were collected using a laser scanning confocal microscope. The collection process included spectral scanning with an excitation wavelength of 488 nm, a scanning range of 500 nm to 794 nm, and a step length of 3 nm, to collect absorption spectrum data of each sample;

[0011] (2) Data preprocessing: spectral data processing, including data set division, baseline correction, data smoothing and standardization;

[0012] a. Dataset division: The spectral dataset is divided into two parts: the calibration set is used for model training, accounting for 70% of the total data; the prediction set is used for model verification, accounting for 30% of the total data; the dataset division adopts the Kennard Stone algorithm;

[0013] b. Baseline correction: Baseline correction is performed on the original spectral data to eliminate low-frequency drift caused by instrument noise or changes in experimental conditions. Baseline correction is achieved through a first-order polynomial fitting method;

[0014] c. Data smoothing: By calculating the mean of data points within a certain window, high-frequency noise is smoothed and removed, thereby improving the stability of the data. The smoothing operation is achieved through convolution;

[0015] d. Standard normal variable transformation was adopted, that is, the mean of each data point was subtracted and divided by the standard deviation, so as to transform all spectral data into a standard normal distribution with a mean of 0 and a standard deviation of 1;

[0016] (3) Feature screening: The Pearson correlation coefficient was used to screen the spectral features and the top 20 features with the strongest correlation with crude oil density were selected;

[0017] (4) Model establishment: The support vector machine regression (SVR) model was selected as the prediction model, which uses the kernel function to model nonlinear relationships in high-dimensional space. The core parameters of the model include the regularization coefficient C, epsilon∈, the kernel function type, and the kernel function coefficient γ. In order to optimize the model performance, preliminary hyperparameter tuning was first performed, and a smaller search grid was set, where C∈{1, 10}, ∈∈{0.01, 0.1}, the kernel function selected the radial basis function RBF, and γ was set to scale. The optimal parameters were found on the training set through grid search GridSearchCV combined with 3-fold cross validation. Based on the preliminary tuned parameters: C=3, ∈=0.02, the SVR model was trained and the performance was evaluated on the training set and the test set.

[0018] (5) Model validation: Evaluation indicators include root mean square error (RMSE) and coefficient of determination (R) 2 , verify the prediction accuracy of the established model on the test set and calculate the prediction error;

[0019] (6) Model application: Using the interactive software, users can upload the spectral data of crude oil in the rock slice to be tested, and use the software to generate prediction results and visualization images of crude oil density.

[0020] In the above step (1), the spectral data is collected using an XYλ scanning mode, and 99 absorption spectrum data points are collected for each sample.

[0021] In the above step (2):

[0022] The specific operation of baseline correction is as follows:

[0023] For each spectral data yi, a linear polynomial is fitted by the least squares method:

[0024] y baseline=a·x+b

[0025] Among them, a and b are the coefficients obtained by fitting, x is the index of the data point, and y is baseline is the fitted baseline;

[0026] Subtract the fitted baseline from the raw spectral data to obtain the corrected data:

[0027] y corrected (i) = y(i) - y baseline (i)

[0028] Among them, y corrected (i) is the spectral data after baseline correction;

[0029] The specific operations of data smoothing are as follows:

[0030] For the i-th data point, its smoothed value y smoothed (i) is:

[0031]

[0032] Where N is the window size, which is usually selected to be 3 to 5 data points. This operation achieves smoothing by applying a convolution kernel to the data points.

[0033] The specific formula for standardization is as follows:

[0034] For the i-th data point, y(i) is its standardized value y snv (i) is:

[0035]

[0036] Where μ is the mean of the spectral data, and the calculation formula is:

[0037]

[0038] σ is the standard deviation of the spectral data, calculated as:

[0039]

[0040] Where M is the total number of spectral data points. Through this normalization process, all data points are converted to a standard normal distribution with zero mean and unit variance.

[0041] In the above step (3):

[0042] The Pearson correlation coefficient formula is as follows:

[0043]

[0044] Where r is the Pearson correlation coefficient; the xi variable is the i-th data point of x; y i is the i-th data point of variable y; is the mean of variable X, is the mean of variable y; ∑: summation symbol.

[0045] In the above step (4), the kernel function of the support vector machine regression model is a radial basis function RBF, and the optimization parameters include a regularization coefficient C and kernel function parameters.

[0046] In the above step (5), the model verification results are expressed as root mean square error RMSE and determination coefficient R 2 The results of the measurement and verification show that R 2 is 0.9558;

[0047] The root mean square error (RMSE) and the coefficient of determination (R2) were calculated using the following formulas:

[0048]

[0049] Where: n is the number of data points (total number of samples), y i is the true value (actual observation value) of the i-th sample, is the predicted value of the i-th sample (model prediction result), is the mean of the data points.

[0050] In the above step (6), the interactive software supports input data formats including CSV, Excel and JSON, and the generated analysis results can be exported as PDF files.

[0051] The above-mentioned software interface includes a data uploading module, a spectral preprocessing module, a result visualization module and a report exporting module; the data preprocessing module is used for baseline correction, data smoothing and standardization processing; the feature screening module is used to screen the spectral features with the strongest correlation with crude oil density; the model prediction module is used to predict crude oil density based on the SVM regression model; the visualization module is used to generate images and analysis reports of crude oil density prediction results.

[0052] The beneficial effects of the present invention are as follows: This application focuses on solving the problems of poor adaptability, long time consumption and low prediction accuracy of traditional methods in complex samples. Spectral data is collected through a confocal microscope, and a support vector machine regression model is combined to achieve rapid prediction of crude oil density. Interactive software is used to provide real-time analysis, visualization and report generation functions of density data, providing a scientific basis for optimizing thermal drive conditions and improving the efficiency of crude oil quality assessment in oil field development. The method should have ease of operation, high-precision prediction capabilities and a wide range of application scenarios to meet the needs of the modern petroleum industry for rapid assessment, dynamic monitoring and optimized decision-making; a crude oil density detection method that is efficient, accurate, highly applicable and has real-time visualization functions. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a correlation diagram of the original signal of the crude oil spectrum measured by CLSM in the present invention;

[0054] Figure 2 The calibration set and the test set correlation diagram divided by the Kennard Stone algorithm of the present invention;

[0055] Figure 3 This is a correlation diagram of crude oil spectrum after baseline correction measured by CLSM in the present invention;

[0056] Figure 4 This is a correlation diagram of the smoothed crude oil spectrum measured by CLSM in the present invention;

[0057] Figure 5 This is a correlation diagram of the crude oil spectrum after normalization measured by CLSM in the present invention;

[0058] Figure 6 The present invention uses the Pearson coefficient to calculate the correlation diagram of the 20 spectral signals with the strongest density correlation;

[0059] Figure 7 It is a correlation diagram between the density prediction value and the density standard value of the present invention;

[0060] Figure 8 This is a low temperature sealing correlation diagram of the massive mudstone core sampled in the present invention;

[0061] Fig. 9 This is a correlation diagram of the blocky mudstone thin section sampled in the present invention;

[0062] Fig.10 This is a density visualization interactive interface related diagram developed by the present invention based on the SVMR model;

[0063] Fig.11 The crude oil density visualization correlation diagram of the rock thin section at 50°C scanned by CLSM for the present invention;

[0064] Fig.12 The crude oil density visualization correlation diagram of the rock thin section at 70°C scanned by CLSM for the present invention;

[0065] Fig.13 Crude oil density visualization correlation diagram of rock slices scanned at 90°C using CLSM for the present invention. DETAILED DESCRIPTION

[0066] The present invention is further described below in conjunction with experiments:

[0067] A method for detecting and visualizing crude oil density in rock thin sections based on CLSM fluorescence spectroscopy comprises the following steps:

[0068] Step 1: Data Collection

[0069] In this experiment, a Leica SP8 confocal microscope was used to collect spectral data of crude oil samples. 33 bottles of crude oil samples with different densities were selected, with a density range of 0.764 g / cm 3 To 0.9655g / cm 3 (See Table 1). The scanning parameters of the experimental settings are: excitation wavelength of 488 nm, scanning mode of XYλ mode, scanning wavelength range of 500 nm to 794 nm, wavelength step of 3 nm, and 99 absorption spectrum data points for each sample (see Figure 1 ). The detector type is photomultiplier tube (PMT).

[0070] Table 1 Standard values ​​of crude oil density

[0071]

[0072] Step 2: Data Preprocessing

[0073] During the experiment, the spectral data is first preprocessed to ensure the quality of the data and the reliability of subsequent analysis. This step includes the following sub-steps:

[0074] (1) Dataset division: The spectral dataset is divided into two parts: the calibration set and the prediction set (see Figure 2 ). The calibration set is used for model training, accounting for 70% of the total data; the prediction set is used for model validation, accounting for 30% of the total data. The data set is divided using the Kennard Stone algorithm, which calculates the distance between samples and selects the most representative samples in the feature space to ensure that the calibration set and the prediction set have a uniform distribution in the feature space, thereby improving the generalization ability of the model.

[0075] (2) Baseline correction: Baseline correction is performed on the raw spectral data to eliminate low-frequency drift caused by instrument noise or changes in experimental conditions. Baseline correction is achieved by a first-order polynomial fitting method. The fitted baseline is subtracted from the raw spectral data to obtain the baseline-corrected spectral data (see Figure 3 ). The specific operations are as follows:

[0076] For each spectral data y( i) , fit a first-degree polynomial by the least squares method:

[0077] y baseline =a·x+b

[0078] Among them, a and b are the coefficients obtained by fitting, x is the index of the data point, and y is baseline is the fitted baseline.

[0079] Subtract the fitted baseline from the raw spectral data to obtain the corrected data:

[0080] y corrected (i) = y(i) - y baseline (i)

[0081] Among them, y corrected (i) is the spectral data after baseline correction.

[0082] (3) Data smoothing: In order to reduce the noise in the spectral data, the moving average method is used to smooth the corrected data. This method smoothes and removes high-frequency noise by calculating the mean of the data points within a certain window, thereby improving the stability of the data. The smoothing operation is achieved through convolution, specifically: for the i-th data point, its smoothed value y smoothed (i) is:

[0083]

[0084] Where N is the width of the smoothing window, which indicates the number of adjacent data points considered each time the mean is calculated; i is the index of the data point currently being smoothed; j is the index of the data point in the window currently involved in the calculation, and y(j) is the value of the jth original data point. (See Figure 4 ).

[0085] (4) Standardization: The smoothed data is standardized using the standard normal variable transformation (SNV), that is, each data point is processed by subtracting the mean and dividing by the standard deviation to transform all spectral data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The specific formula is as follows:

[0086] For the i-th data point y i , its standardized value y snv (i) is:

[0087]

[0088] Where μ is the mean of the spectral data, and the calculation formula is:

[0089]

[0090] σ is the standard deviation of the spectral data, calculated as:

[0091]

[0092] Where M is the total number of spectral data points. Through this standardization process, all data points are converted to a standard normal distribution with zero mean and unit variance. The standardized spectral data (see Figure 5 ).

[0093] Step 3: Feature Screening

[0094] In order to improve the training efficiency of the model and reduce the impact of redundant features on prediction performance, the following steps are used to screen important features:

[0095] Correlation analysis: Calculate the Pearson correlation coefficient between each spectral feature and the target variable Density.

[0096] The correlation coefficient formula is as follows:

[0097]

[0098] Where r is the Pearson correlation coefficient; x i The variable is the i-th data point of x; y i is the i-th data point of variable y; is the mean of the variable X. is the mean value of variable y; ∑: summation symbol; filtered spectral data (see Table 2).

[0099] The results show that the higher the correlation coefficient value, the stronger the feature's ability to explain the target variable. The spectral features are sorted from high to low according to the absolute value of the correlation, and the top 20 features most correlated with the density value are selected as the model input features (see Figure 6 ).

[0100] Table 2 The top 20 spectral signals with the strongest correlation with density

[0101]

[0102] Step 4: SVMR model establishment:

[0103] The support vector machine regression (SVR) model was selected as the prediction model, which uses kernel functions to model nonlinear relationships in high-dimensional space. The core parameters of the model include the regularization coefficient C, epsilon∈, the kernel function type, and the kernel function coefficient γ. In order to optimize the performance of the model, a preliminary hyperparameter tuning was first performed, and a smaller search grid was set, where C∈{1, 10}, ∈∈{0.01, 0.1}, the kernel function selected radial basis function (RBF), and γ was set to scale. The optimal parameters were found on the training set through grid search (GridSearchCV) combined with 3-fold cross validation. Based on the preliminary tuned parameters: C=3, ∈=0.02, the SVR model was trained and the performance was evaluated on the training set and the test set.

[0104] Step 5: Model Validation

[0105] Evaluation indicators include root mean square error (RMSE) and determination coefficient (R 2 ), and are calculated using the following formulas:

[0106]

[0107] Where: n is the number of data points (total number of samples), yi is the true value (actual observation value) of the ith sample, is the predicted value of the i-th sample (model prediction result), is the mean of the data points.

[0108] After the prediction data is statistically analyzed by the above formula, the correlation coefficient of the SVM regression model is obtained (see Figure 7 , Table 3) and the error of the test sample (see Table 4).

[0109] Table 3 Comparison of predicted density and standard density of test set

[0110]

[0111] Table 4 Prediction density error analysis of the test set

[0112]

[0113] Step 6: Model Application

[0114] On the basis of establishing the SVM regression model and confirming that it has good fitting accuracy (R square = 0.9558), the samples were collected from the Qingshankou Formation in the Kulong Sag of the Songliao Basin, with a depth of 3200 to 3220 meters. The core is massive mudstone with low porosity (about 0.5mD) and oil saturation of 40%. After the sample was collected, the core was stored in a sealed container and frozen at -20℃ for 24 hours (see Figure 8). This low temperature treatment effectively prevents the loss of light hydrocarbons and ensures that the sample remains stable during subsequent processing.

[0115] The frozen samples were then precisely cut into 1 cm 3 After cutting, the samples were dried at low temperature and bonded using α-cyanoacrylate in a vacuum environment to ensure that there was no air bubble interference. After rough grinding, fine grinding and polishing, the final sample thickness was reduced to 0.04-0.05 mm, and then the sample was placed on the CLSM stage (see Fig. 9 ), using the same scanning parameters and steps as for crude oil, the fluorescence spectrum files of rock slices heated to 50℃, 70℃, and 90℃ were obtained.

[0116] The support vector machine (SVM) regression algorithm of Python code is used to analyze and predict the characteristics of spectral data. The fluorescence spectral image analysis system uses Python's Tkinter library to build a graphical user interface (GUI) and combines Matplotlib to achieve data visualization (see Fig.10 ). The interface design is divided into two main areas: the left side is used for file selection, data processing startup and progress display, and the right side is used for image display. Users can upload single or paired spectral data files through the interface, supporting CSV, Excel, JSON and binary formats. The system automatically reads the data, performs normalization and spectral analysis, and generates crude oil density in rock slices at 50℃ (see Fig.11 )、70℃(see Fig.12 )、90℃(see Fig.13 ) visualization.

Claims

1. A method for detecting and visualizing crude oil density in rock thin sections based on CLSM fluorescence spectroscopy, comprising the following steps: (1) Data collection: spectral data of crude oil samples were collected using a laser scanning confocal microscope. The collection process included spectral scanning with an excitation wavelength of 488 nm, a scanning range of 500 nm to 794 nm, and a step length of 3 nm, to collect absorption spectrum data of each sample; (2) Data preprocessing: spectral data processing, including data set division, baseline correction, data smoothing and standardization; a. Dataset division: The spectral dataset is divided into two parts: the calibration set is used for model training, accounting for 70% of the total data; the prediction set is used for model verification, accounting for 30% of the total data; the dataset division adopts the Kennard Stone algorithm; b. Baseline correction: Baseline correction is performed on the original spectral data to eliminate low-frequency drift caused by instrument noise or changes in experimental conditions. Baseline correction is achieved through a first-order polynomial fitting method; c. Data smoothing: By calculating the mean of data points within a certain window, high-frequency noise is smoothed and removed, thereby improving the stability of the data. The smoothing operation is achieved through convolution; d. Standard normal variable transformation was adopted, that is, the mean of each data point was subtracted and divided by the standard deviation, so as to transform all spectral data into a standard normal distribution with a mean of 0 and a standard deviation of 1; (3) Feature screening: The Pearson correlation coefficient was used to screen the spectral features and the top 20 features with the strongest correlation with crude oil density were selected; (4) Model establishment: The support vector machine regression (SVR) model was selected as the prediction model, which uses the kernel function to model nonlinear relationships in high-dimensional space. The core parameters of the model include the regularization coefficient C, epsilon∈, the kernel function type, and the kernel function coefficient γ. In order to optimize the model performance, preliminary hyperparameter tuning was first performed, and a smaller search grid was set, where C∈{1, 10}, ∈∈{0.01, 0.1}, the kernel function selected the radial basis function RBF, and γ was set to scale. The optimal parameters were found on the training set through grid search GridSearchCV combined with 3-fold cross validation. Based on the preliminary tuned parameters: C=3, ∈=0.02, the SVR model was trained and the performance was evaluated on the training set and the test set. (5) Model validation: Evaluation indicators include root mean square error (RMSE) and coefficient of determination (R) 2 , verify the prediction accuracy of the established model on the test set and calculate the prediction error; (6) Model application: Using the interactive software, users can upload the spectral data of crude oil in the rock slice to be tested, and use the software to generate prediction results and visualization images of crude oil density.

2. The method for detecting and visualizing crude oil density in rock slices based on CLSM fluorescence spectroscopy according to claim 1, characterized in that: In step (1), the spectral data is collected using an XYλ scanning mode, and 99 absorption spectrum data points are collected for each sample.

3. The method for detecting and visualizing crude oil density in rock slices based on CLSM fluorescence spectroscopy according to claim 1, characterized in that: In step (2): The specific operation of baseline correction is as follows: For each spectral data yi, a linear polynomial is fitted by the least squares method: y baseline =a·x+b Among them, a and b are the coefficients obtained by fitting, x is the index of the data point, and y is baseline is the fitted baseline; Subtract the fitted baseline from the raw spectral data to obtain the corrected data: y corrected (i)=y(i)-y baseline (i) Among them, y corrected (i) is the spectral data after baseline correction; The specific operations of data smoothing are as follows: For the i-th data point, its smoothed value y smoothed (i) is: Where N is the window size, which is usually selected to be 3 to 5 data points. This operation achieves smoothing by applying a convolution kernel to the data points. The specific formula for standardization is as follows: For the i-th data point y i , its standardized value y snv (i) is: Where μ is the mean of the spectral data, and the calculation formula is: σ is the standard deviation of the spectral data, calculated as: Where M is the total number of spectral data points; through this normalization process, all data points are converted to a standard normal distribution with zero mean and unit variance.

4. The method for detecting and visualizing crude oil density in rock slices based on CLSM fluorescence spectroscopy according to claim 1, characterized in that: In step (3): The Pearson correlation coefficient formula is as follows: Where r is the Pearson correlation coefficient; the xi variable is the i-th data point of x; yi is the i-th data point of the variable y; is the mean of variable X, is the mean of variable y; ∑: summation symbol.

5. The method for detecting and visualizing crude oil density in rock slices based on CLSM fluorescence spectroscopy according to claim 1, characterized in that: In step (4), the kernel function of the support vector machine regression model is a radial basis function RBF, and the optimization parameters include a regularization coefficient C and kernel function parameters.

6. The method for detecting and visualizing crude oil density in rock slices based on CLSM fluorescence spectroscopy according to claim 1, characterized in that: In step (5), the model validation results are expressed as root mean square error RMSE and determination coefficient R 2 The results of the measurement and verification show that R 2 is 0.9558; The root mean square error (RMSE) and the coefficient of determination (R2) were calculated using the following formulas: Where: n is the number of data points (total number of samples), y i is the true value (actual observation value) of the i-th sample, is the predicted value of the i-th sample (model prediction result), is the mean of the data points.

7. The method for detecting and visualizing crude oil density in rock slices based on CLSM fluorescence spectroscopy according to claim 1, characterized in that: In step (6), the interactive software supports input data formats including CSV, Excel and JSON, and the generated analysis results can be exported as PDF files.

Images