Hydrocarbon source physical property interpretation system and method based on infrared spectrum absorption characteristics

Abstract

The invention relates to the technical field of hydrocarbon source rock infrared spectrum analysis, in particular to a hydrocarbon source physical property interpretation system and method based on infrared spectrum absorption characteristics, and the system comprises a spectrum data acquisition module, a characteristic separation module, a physical property mapping module and a geological interpretation module. The feature separation module controls a convolutional neural network to decouple a mineral interference absorption peak through a second derivative spectrum saddle point threshold, a dynamic interference database is combined with a transfer learning mechanism to adapt to regional mineral combination changes, and organic functional group separation features with clay interference labels and half-peak width quantized values are output. The physical property mapping module activates horizon specificity correction based on three-dimensional coordinates and two-factor verification, and dynamically adjusts inversion parameters of total organic carbon and hydrogen indexes. And the geological interpretation module updates a mineral template threshold according to a hydrocarbon yield inversion result, particle swarm optimization algorithm output drives convolutional neural network weight cross-cycle iteration, and a closed-loop optimization system for cooperative enhancement of spectral decoupling precision and geological model convergence is formed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of infrared spectrum analysis of hydrocarbon source rock, and in particular to a hydrocarbon source property interpretation system and method based on infrared spectrum absorption characteristics. BACKGROUND

[0002] The hydrocarbon source property interpretation system is a comprehensive platform for evaluating the potential of hydrocarbon source rock in oil and gas geological exploration. The system integrates rock geochemical analysis and geophysical measurement data to analyze total organic carbon content, thermal maturity indicators and hydrocarbon generation kinetics parameters of hydrocarbon source rock, so as to infer the conversion degree of organic matter in rock and the oil and gas generation capacity. The logical basis is that the abundance of organic matter and thermal history directly affect the scale and quality of oil and gas resources, so the system can build a geological model and predict resource distribution to assist exploration personnel in formulating an optimized scheme to reduce development risks and improve resource utilization.

[0003] The technical pain point of the hydrocarbon source property interpretation system based on infrared spectrum absorption characteristics in the aspect of analysis is the qualitative and quantitative uncertainty introduced by the superposition of spectrum absorption characteristics. The core logic is that the inversion of infrared spectrum data depends on the accurate identification of functional group absorption peak position and intensity, but the silicate and carbonate inorganic mineral components contained in the hydrocarbon source rock sample can produce overlapping absorption signals in a specific wave band, form background interference and shield the spectral response of organic functional groups, resulting in deviation in the inference of organic matter abundance and type. For example, when analyzing shale containing high clay minerals, the broadband absorption of interlayer water molecules of clay at about 3400 cm-1 is easy to be confused with the peak of organic oxygen-containing groups, which can incorrectly increase the thermal maturity index value or underestimate the actual value of total organic carbon content, ultimately weakening the prediction reliability of the hydrocarbon generation kinetics model and misleading the resource distribution evaluation. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a hydrocarbon source property interpretation system and method based on infrared spectrum absorption characteristics, which solves the problem of deviation in qualitative and quantitative inference of organic matter abundance and type caused by the superposition of absorption characteristics of inorganic mineral components in the infrared spectrum analysis of hydrocarbon source rock.

[0005] To solve the above technical problems, the present application is as follows: In a first aspect, the present application provides a hydrocarbon source property interpretation system based on infrared spectrum absorption characteristics, which comprises a spectrum data acquisition module, a feature separation module, a property mapping module and a geological interpretation module. The spectrum data acquisition module is configured to acquire in-situ gas infrared spectrum and core micro-area infrared spectrum of the hydrocarbon source rock sample. The feature separation module is configured to: detect the second derivative spectrum saddle point in the interval of 1700-1800 cm-1; ​When the second derivative spectrum saddle point is greater than or equal to 3, a convolutional neural network is started to decouple overlapping absorption peaks; At 3400 When the wave band half peak width is greater than 120 The clay interference label is activated; The organic functional group separation feature with the clay interference label is output; The physical property mapping module is configured to be configured to: The separation feature with the clay interference label is received; When the aromatic ring absorption peak symmetry and the fat chain peak intensity ratio difference are greater than 20%, the clay correction factor is triggered; The corrected total organic carbon content, thermal maturity and hydrogen index parameters are output; The geological interpretation module is configured to be configured to: When the hydrocarbon activation energy dispersion coefficient is greater than 0.5, a particle swarm optimization algorithm is started to invert the hydrocarbon yield curve; The mineral anomaly area distribution data is received and the silicate template matching threshold is updated to 85%. The hydrocarbon yield optimal solution output by the particle swarm optimization algorithm corrects the convolutional neural network weight parameters, and the updated weight parameters are input into the feature separation module for next period decoupling calculation.

[0006] Further, the hydrocarbon source physical property interpretation system based on infrared spectrum absorption characteristics provided by the application comprises: The in-situ gas collection unit separates hydrocarbon gas through a semi-permeable membrane separator at the outlet of the drilling fluid, and inputs the gas into a gas absorption cell after removing carbonate dust through a molecular sieve, to generate a 1300-3500 nm wave band time spectrum; The rock sample collection unit polishes the core sample with argon ions and scans on a micro-infrared platform to output a transmission and reflection dual-mode spectrum matrix of 50 mu m x 50 mu m micro area; the feature separation module receives the time spectrum and the dual-mode spectrum matrix, and establishes a three-dimensional mapping relationship according to the well depth coordinates.

[0007] Further, the hydrocarbon source physical property interpretation system based on infrared spectrum absorption characteristics provided by the application comprises: The dynamic interference database stores silicate 1100-1000 And carbonate 1500-1400 Wave band feature template, when the wave band feature template matching degree is less than 90%, the transfer learning mechanism is activated; The spectrum peak decoupling unit calculates 1700- Interval second derivative spectrum saddle point: When the second derivative spectrum saddle point is greater than or equal to 3, a residual connection convolutional neural network is started, and the original spectrum and the second derivative spectrum are input in parallel data; The organic functional group separation feature with three-dimensional coordinates and clay interference label of the output band is output to the physical property mapping module.

[0008] Further, the hydrocarbon source physical property interpretation system based on infrared spectrum absorption characteristics provided by the present application, the spectral peak decoupling unit is configured to: At 3400 The half-peak width of the wave band is greater than 120 When the clay interference label is activated, the half-peak width quantization value is recorded; The quantization value is synchronously transmitted to the physical property mapping module with the decoupled spectrum data with three-dimensional coordinates; Wherein, the clay interference label is directly related to the total organic carbon compensation mechanism of the physical property mapping module, and when the half-peak width quantization value is greater than 130 The compensation coefficient is increased to 1.2 times.

[0009] Further, the hydrocarbon source physical property interpretation system based on infrared spectrum absorption characteristics provided by the present application, the physical property mapping module is configured to: Receive decoupled spectrum data carrying three-dimensional coordinates and half-peak width quantization value; When the aromatic ring absorption peak symmetry index is less than 0.85 and the difference between the peak intensity ratio of the aliphatic chain is greater than 20%, the clay correction factor is activated; Convert the alkyl / carboxyl peak area ratio to hydrogen index by pre-training the regression model; Wherein, the half-peak width quantization value dynamically adjusts the temperature correction parameter of the regression model, and the half-peak width quantization value increases by 10 , the temperature correction slope is increased by 0.15.

[0010] Further, the hydrocarbon source physical property interpretation system based on infrared spectrum absorption characteristics provided by the present application, the geological interpretation module is configured to: When the thermal maturity parameter changes by more than a set scale in the vertical depth gradient, activate the fault structure analysis protocol; Receive the hydrogen index parameter transmitted by the physical property mapping module, and when the product of the hydrogen index and the discrete coefficient of the hydrocarbon activation energy reaches the trigger threshold, start the particle swarm optimization algorithm to perform multi-objective parameter inversion; Wherein, the initial population size of the particle swarm optimization algorithm is constrained by the half-peak width quantization value output by the feature separation module.

[0011] Further, the hydrocarbon source physical property interpretation system based on infrared spectrum absorption characteristics provided by the present application, the feedback channel of the geological interpretation module and the feature separation module is configured to: Analyze the carbonate enrichment layer distribution information in the three-dimensional quality grading result; When the carbonate enrichment ratio is higher than the critical value and the silicate template matching degree is lower than the current threshold: Activate the transfer learning mechanism of the dynamic interference database; The weight parameters of the convolutional neural network are updated based on the hydrocarbon yield inversion residuals; The updated weight parameters are input into the spectral peak decoupling unit to perform subsequent decoupling calculations.

[0012] Furthermore, in the hydrocarbon source property interpretation system based on infrared spectral absorption characteristics described in this invention, the data transmission channel between the feature separation module and the property mapping module is configured as follows: When the decoupled spectral data carries a clay interference tag, a verification trigger command is sent to the property mapping module. The thermal maturity verification unit of the physical property mapping module activates the clay correction factor calculation protocol according to the instruction. The transmitted data includes the half-peak width quantization value and three-dimensional coordinate information, and the trigger command and clay interference label are generated simultaneously.

[0013] Furthermore, in the hydrocarbon source property interpretation system based on infrared spectral absorption characteristics described in this invention, the closed-loop channel between the geological interpretation module and the feature separation module is configured as follows: After the particle swarm optimization algorithm outputs the optimal solution for hydrocarbon yield, the rate of change of residual in the current decoupling cycle is calculated. When the rate of change of the residual exceeds the set range, the weight parameters of the fully connected layer of the convolutional neural network are adjusted. The updated weight parameters are immediately input into the feature separation module to start the next round of saddle point detection iteration; Among them, the weight correction amount is negatively correlated with the goodness of fit of the hydrocarbon yield curve.

[0014] Secondly, the hydrocarbon source property interpretation method based on infrared spectral absorption characteristics provided by the present invention, applied to the aforementioned hydrocarbon source property interpretation method based on infrared spectral absorption characteristics, includes: Step 1: Collect in-situ gas infrared spectra and core micro-area infrared spectra of source rocks to generate time-series spectra in the 1300-3500nm band, and obtain the transmission mode spectral matrix and reflection mode spectral matrix of 50μm micro-area. Step 2, Identify mineral interference and decouple organic features: Detect 1700- The second derivative spectrum saddle points in the interval are used to initiate a convolutional neural network to decouple overlapping peaks when the number of saddle points is ≥3, and at 3400 Band half-width > 120 The clay interference label is activated at time, and the full width at half maximum (FWHM) quantization value is recorded. Step 3, perform dynamic correction of physical property parameters: receive the separation characteristics with clay interference tags and half-peak width quantization values, and trigger the clay correction factor when the symmetry index of the aromatic ring absorption peak is <0.85 and the difference in peak intensity ratio of the aliphatic chain is >20%. Step 4, build a geological resource grading model: when the hydrocarbon activation energy dispersion coefficient is greater than 0.5, start the particle swarm optimization algorithm to inverse the hydrocarbon yield curve, and update the silicate template matching threshold to 85% based on the mineral anomaly area distribution data; Step 5, correct the convolutional neural network weight parameters according to the optimal solution of the hydrocarbon yield, input the updated weight parameters into the next period of decoupling calculation, and return to step 2 to perform iteration.

[0015] Advantages of the present application The present application breaks through the technical limitations of the recognition deviation of organic functional groups caused by the superposition of inorganic minerals in traditional spectral analysis through the synergistic effect of dynamic decoupling of mineral interference absorption characteristics, layer-specific parameter correction and closed-loop feedback optimization: the saddle point triggering mechanism of the characteristic separation module dynamically adapts to the change of regional mineral combination in combination with transfer learning, and accurately removes the absorption interference of silicates and carbonates in a specific waveband; the physical property mapping module performs layer adaptive compensation based on three-dimensional coordinates and quantitative labels, eliminating the systematic error of the homogenization model on the quantitative inversion of organic matter abundance; the geological interpretation module drives the network weight to iterate across periods according to the residual error of the hydrocarbon yield inversion, forming a technical closed loop with continuously enhanced mineral interference suppression capability, and finally realizing the synergistic optimization of the qualitative and quantitative parameter inference precision of organic matter type. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the present application, the following will briefly introduce the drawings needed in the embodiments of the present application. Obviously, other drawings can also be obtained by those skilled in the art without creative labor.

[0017] Figure 1 The flowchart of the hydrocarbon source physical property interpretation method based on infrared spectral absorption characteristics provided by the embodiments of the present application. DETAILED DESCRIPTION

[0018] In order to make the purpose, the present application and the advantages of the present application clearer, the present application will be described clearly and completely in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application. The present application provided by the embodiments of the present application will be described in combination with the drawings. In order to better understand the purpose of the present application, the present application will be further described in detail.

[0019] In the first aspect, the present application provides a hydrocarbon source physical property interpretation system based on infrared spectral absorption characteristics, which comprises a spectral data acquisition module, a characteristic separation module, a physical property mapping module and a geological interpretation module. The spectrum data acquisition module is configured to acquire in-situ gas infrared spectrum and core micro-area infrared spectrum of the hydrocarbon source rock sample. The feature separation module is configured to be configured to: Detection 1700- Second derivative spectrum saddle point of the interval; When the second derivative spectrum saddle point is greater than or equal to 3, a convolutional neural network is started to decouple overlapping absorption peaks; At 3400 When the wave band half peak width is greater than 120 The clay interference label is activated; Output organic functional group separation features with clay interference labels; The physical property mapping module is configured to be configured to: Receive separation features with clay interference labels; When the difference between the aromatic ring absorption peak symmetry and the aliphatic chain peak intensity ratio is greater than 20%, a clay correction factor is triggered; Output corrected total organic carbon content, thermal maturity and hydrogen index parameters; The geological interpretation module is configured to be configured to: When the hydrocarbon activation energy dispersion coefficient is greater than 0.5, a particle swarm optimization algorithm is started to invert the hydrocarbon yield curve; Receive mineral anomaly area distribution data and update the silicate template matching threshold to 85%; The particle swarm optimization algorithm outputs the optimal solution of the hydrocarbon yield, which corrects the convolutional neural network weight parameters, and the updated weight parameters are input into the feature separation module for next period of decoupling calculation.

[0020] The spectrum data acquisition module performs in-situ gas infrared spectrum and core micro-area infrared spectrum acquisition operation of the hydrocarbon source rock sample. The in-situ gas acquisition unit is disposed with a semi-permeable membrane separator at the outlet of the drilling fluid, separates hydrocarbon gas, filters carbonate dust through molecular sieve, and generates time sequence spectrum of specific wave band in the gas absorption cell; the rock sample acquisition unit ionizes and polishes the core sample, and the micro-infrared platform scans to output micro-area transmission mode and reflection mode spectrum matrix. The spectrum data obtained by the module has spatial resolution characteristics.

[0021] The feature separation module receives the above-mentioned spectral data and performs interference recognition and feature decoupling operations. The dynamic interference database stores inorganic mineral wave band feature templates, and the migration learning mechanism is activated when the template matching degree is insufficient to load adjacent area data. The spectral peak decoupling unit calculates the second derivative spectral saddle point of the specified interval, and when the number of saddle points reaches a set threshold, the convolutional neural network with a residual connection structure is started to process the original spectrum and derivative spectrum parallel data to decouple the organic-inorganic overlapping absorption peaks. The clay interference label is activated and the quantitative value is recorded when detecting the half-peak width anomaly in the specific hydroxyl absorption wave band, and the organic functional group separation feature with three-dimensional coordinate label is output. The module controls the start and stop of the neural network through the second derivative spectral saddle point threshold, and realizes dynamic interference suppression by combining migration learning.

[0022] The physical property mapping module receives the decoupled spectral data with labels and performs parameter conversion. When the aromatic ring absorption peak symmetry index and the fat chain peak intensity ratio exceed the set threshold at the same time, the clay correction factor is activated. The pre-trained regression model converts the peak area ratio of alkyl peaks and carboxyl peaks into hydrogen index parameters, and the half-peak width quantitative value dynamically adjusts the temperature correction parameter slope of the model. The total organic carbon content compensation mechanism is directly related to the clay interference label. The module locks the mineral interference type through a double-factor verification mechanism to realize dynamic correction of parameters.

[0023] The geological interpretation module constructs a resource evaluation model based on the physical property parameters. When the hydrocarbon generation activation energy dispersion coefficient exceeds the limit, the particle swarm optimization algorithm starts the hydrocarbon yield curve inversion; the reservoir analysis unit triggers the simulated annealing algorithm to optimize the three-dimensional quality classification according to the permeability variation coefficient. The distribution data of mineral abnormal areas drive the update of the silicate template matching threshold. The optimal solution of the hydrocarbon yield output by the particle swarm algorithm is used to correct the convolutional neural network weight parameters.

[0024] The system sets a closed-loop feedback channel: the three-dimensional quality classification results output by the geological interpretation module are analyzed, and the migration learning mechanism of the dynamic interference database is activated for the carbonate enrichment over-limit horizon; the convolutional neural network weight parameters are updated according to the residual error of the hydrocarbon yield inversion; the updated weight parameters return to the feature separation module to start the next round of decoupling calculation. This feedback mechanism makes the mineral interference recognition accuracy continue to evolve with the number of iterations.

[0025] A directional data transmission channel is established between each module: when the feature separation module transmits decoupled data to the physical property mapping module, the clay interference label synchronously sends a verification trigger instruction; the physical property mapping module activates the correction factor calculation protocol according to the instruction. The geological interpretation module and the feature separation module form a cross-cycle optimization loop, the weight correction amount is dynamically adjusted by the fitting degree of the hydrocarbon yield curve, and the self-improvement of the system processing efficiency is realized.

[0026] Specifically, the hydrocarbon source physical property interpretation system based on infrared spectral absorption characteristics provided by the present application comprises: In-situ gas collection unit, separating hydrocarbon gas at the outlet of drilling fluid through a semi-permeable membrane separator, inputting the gas into a gas absorption cell after filtering out carbonate dust through a molecular sieve, and generating time-series spectra in the 1300-3500 nm waveband; Rock sample collection unit, polishing the core sample with argon ion and scanning on a micro-infrared platform, outputting a 50 μm x 50 μm micro-area transmission and reflection dual-mode spectral matrix; The feature separation module receives the time-series spectra and the dual-mode spectral matrix, and establishes a three-dimensional mapping relationship according to the well depth coordinates.

[0027] The in-situ gas collection unit deploys a semi-permeable membrane separation device at the outlet of the drilling fluid. After separating the hydrocarbon gas, the device filters out carbonate mineral dust particles through a molecular sieve structure, and the purified gas is input into a gas absorption cell with a specific optical path length. The non-dispersive infrared sensor inside the absorption cell scans a specific waveband range to generate time-series spectral data. This operation eliminates the scattering interference of inorganic solid-phase materials in the drilling fluid on the gas spectrum, ensuring the accuracy of the in-situ hydrocarbon gas absorption characteristics.

[0028] The rock sample collection unit polishes the core sample with an argon ion beam surface polishing process, eliminating the influence of surface oxidation and contaminants on infrared transmission. After polishing, the sample is placed on a micro-infrared spectral platform, and a focal plane array detector scans the micro-area with a specific spatial resolution. The detector simultaneously records the spectral response in transmission and reflection modes, outputting a dual-mode spectral matrix data. This process realizes spatially resolved spectral collection of organic inclusions and adjacent mineral micro-areas.

[0029] The feature separation module receives time-series spectral data from the gas collection unit and dual-mode spectral matrix data from the rock collection unit. The system performs spatial alignment processing on the two types of spectral data according to the drilling depth coordinates, establishing a three-dimensional mapping relationship between wellbore depth, rock lithology, and spectral characteristics. This mapping relationship correlates in-situ gas spectral anomaly sections with core micro-area mineral distribution, providing a spatial positioning reference for subsequent organic-inorganic feature separation.

[0030] The time-series spectral data is labeled with well depth coordinates and input into a dynamic interference database. The database stores standard absorption templates of silicate minerals in the silicon-oxygen bond stretching vibration waveband, as well as characteristic templates of carbonate minerals in the carbon-oxygen anti-symmetric stretching waveband. The rock micro-area spectral matrix data is indexed according to the coordinates and matched with the gas spectrum. When the intensity difference between the gas spectrum and the core spectrum at the same depth position exceeds the set range, the spectral data cross-validation process is triggered.

[0031] The three-dimensional mapping relationship drives the decoupling calculation of the feature separation module. The gas spectrum and the rock spectrum at the same depth coordinate are combined as parallel input data, and the mineral interference template at the corresponding depth is called through coordinate indexing. When the template matching degree is insufficient, the transfer learning mechanism automatically loads the same layer spectrum data of the adjacent exploration well to supplement the training. This design realizes the cooperation of spatial positioning data and spectral processing algorithm, and improves the layer specificity of mineral interference recognition.

[0032] Specifically, the hydrocarbon source property interpretation system based on infrared spectrum absorption characteristics comprises a feature separation module. A dynamic interference database stores silicate 1100-1000 and carbonate 1500-1400 Waveband feature template, when the waveband feature template matching degree < 90%, activate the transfer learning mechanism; Spectral peak decoupling unit, calculate 1700- Interval second derivative spectrum saddle point: When the second derivative spectrum saddle point is greater than or equal to 3, start the residual connection convolutional neural network, input the parallel data of the original spectrum and the second derivative spectrum; Output the organic functional group separation feature with three-dimensional coordinates and clay interference label to the property mapping module.

[0033] The dynamic interference database stores the characteristic absorption template of silicate minerals in the silicon-oxygen bond stretching vibration waveband, and the characteristic absorption template of carbonate minerals in the carbon-oxygen anti-symmetric stretching waveband. When the waveband matching degree of the input spectrum data and the stored template is lower than the set threshold, the transfer learning mechanism is automatically activated. The transfer learning mechanism calls the same layer spectrum data of the adjacent exploration area to train the new mineral combination feature template, and realizes the adaptive recognition of unknown mineral interference.

[0034] The spectral peak decoupling unit performs second derivative transformation processing on the interval of the organic functional group feature peak. The number of second derivative spectrum curvature extreme points is calculated to determine the complexity of the overlapping peak, and when the saddle point number reaches the starting condition, the convolutional neural network with residual connection structure is activated. The network parallelly inputs the original absorption spectrum and the second derivative spectrum data, extracts the frequency domain features through multi-scale convolution kernels, and separates the organic functional group absorption peak under the shielding of silicate and carbonate minerals.

[0035] The feature separation module outputs decoupling spectrum data carrying three-dimensional spatial coordinate information. The three-dimensional coordinates come from the well depth-lithology mapping relationship, and locate the geological layer of the spectrum acquisition point. The decoupling data are also attached with clay mineral interference state labels, which are generated according to the half-peak width anomaly detection results of the hydroxyl stretching vibration waveband. The output data are transmitted to the property mapping module through a structured format, ensuring the complete association of organic functional group features and layer information.

[0036] The migration learning mechanism cooperates with the spectral peak decoupling unit to form a collaborative processing chain. The template matching result output by the dynamic interference database controls the weight initialization parameters of the convolutional neural network. When the migration learning adds a new mineral template, the network automatically loads the pre-trained weights and fine-tunes the parameters of the fully connected layer. This collaborative mechanism enables the decoupling algorithm to dynamically adapt to changes in regional mineral assemblages and eliminate spectral analysis deviations caused by lithological differences.

[0037] The clay interference label generation process integrates spatial positioning data. While detecting abnormalities in the half-peak width of the hydroxyl absorption band, the depth coordinates and lithology codes of the current sampling point are associated. The output label contains stratigraphic attribute information, enabling the subsequent physical property correction module to perform stratigraphic-specific compensation calculations. Label transmission uses the same spatial coordinate system as the decoupled spectrum to maintain spatial consistency in the data chain.

[0038] Specifically, the hydrocarbon source property interpretation system based on infrared spectral absorption characteristics disclosed in the present application, the spectral peak decoupling unit is configured to: At 3400 When the half-peak width is greater than 120 the clay interference label is activated, and the half-peak width quantization value is recorded; The quantization value is transmitted to the physical property mapping module synchronously with the decoupled spectral data with three-dimensional coordinates; Wherein, the clay interference label is directly associated with the total organic carbon compensation mechanism of the physical property mapping module, and when the half-peak width quantization value is greater than 130 the compensation coefficient is increased to 1.2 times.

[0039] The spectral peak decoupling unit performs half-peak width quantization detection on the hydroxyl stretching vibration characteristic band. When the detection value exceeds the set threshold, the clay mineral interference state label is activated and the specific quantization value is recorded. The label generation process associates the three-dimensional coordinate information of the current spectral data to ensure that the interference identification result has geological stratigraphic properties.

[0040] The decoupled organic functional group characteristic data is transmitted together with the clay interference label and the half-peak width quantization value. The data packet is packaged in a structured format, including spatial coordinate index, decoupled spectral matrix, interference state identifier, and quantization parameter. This packaging mechanism ensures that the physical property mapping module can synchronously call multi-dimensional data, avoiding feature information fragmentation.

[0041] The clay interference label is directly associated with the total organic carbon compensation logic of the physical property mapping module. The label activation state triggers the compensation algorithm to start, and the half-peak width quantization value dynamically adjusts the compensation coefficient strength. With each increase in the quantization value by a certain amount, the compensation coefficient is adjusted accordingly, forming a linear response mechanism between interference degree and compensation intensity. This design realizes quantitative compensation control of mineral interference intensity.

[0042] The data transmission process establishes a three-dimensional coordinate driving mechanism. The physical property mapping module matches the geological horizon attribute according to the coordinate index, and generates horizon-specific compensation parameters combined with the half-peak width quantization value. The synergistic effect of spatial coordinates and interference quantization values eliminates the stratigraphic adaptability defects of traditional global compensation models.

[0043] The synchronous transmission of labels and quantization values realizes instant compensation decision. The physical property mapping module receives the decoupled spectrum and obtains the interference state and intensity data at the same time, without the need for additional verification process. This transmission mechanism shortens the mineral interference response delay, ensuring the timeliness and accuracy of total organic carbon content inversion.

[0044] Specifically, the hydrocarbon source physical property interpretation system based on infrared spectrum absorption characteristics provided by the application is configured as: Receiving decoupled spectrum data carrying three-dimensional coordinates and half-peak width quantization values; When the aromatic ring absorption peak symmetry index is less than 0.85 and the difference in aliphatic chain peak intensity ratio is greater than 20%, the clay correction factor is activated; Converting the alkyl / carboxyl peak area ratio to hydrogen index through a pre-trained regression model; Wherein, the half-peak width quantization value dynamically adjusts the temperature correction parameter of the regression model, and the half-peak width quantization value increases by 10 , the temperature correction slope increases by 0.15.

[0045] The physical property mapping module receives the structured data packet transmitted by the feature separation module, which includes three-dimensional spatial coordinate index, decoupled organic functional group spectrum, clay interference state label and half-peak width quantization value. The module matches the geological horizon attribute according to the coordinate index, and establishes the binding relationship between the spectral data and the spatial position.

[0046] The module performs double-parameter verification of the aromatic ring absorption peak symmetry index and the aliphatic chain peak intensity ratio. When the symmetry index is lower than the set threshold and the peak intensity ratio difference exceeds the limited range, the clay correction factor calculation protocol is activated. This double-factor verification mechanism avoids the risk of single-parameter misjudgment and accurately locks the clay mineral interference working condition.

[0047] The pre-trained regression model loads the organic functional group feature data, converts the integral area ratio of alkyl C-H bond stretching vibration peak and carboxyl C=O double bond peak to hydrogen index parameter. The model training uses pyrolysis gas chromatography calibration data set, and establishes the nonlinear mapping relationship between peak area ratio and hydrogen index through supervised learning.

[0048] The half-peak width quantization value dynamically intervenes in the regression model calculation process. The quantization value is input into the temperature correction unit, and the correction function slope parameter is adjusted at a fixed ratio. This adjustment mechanism makes the hydrogen index inversion result adapt to the change of temperature sensitivity characteristics caused by different clay content, and eliminates the system error caused by thermal drift.

[0049] The correction factor activates the injection regression calculation process. The clay correction factor is used as a weight coefficient in the hydrogen index conversion operation to dynamically compensate the alkyl / carboxyl peak area ratio. The compensation strength is linearly related to the half-peak width quantification value, forming an adaptive matching mechanism for the interference degree and correction strength.

[0050] The spatial coordinate index drives the layer-specific correction. The physical property mapping module calls the three-dimensional coordinate-related geothermal gradient data to perform the stratigraphic positioning compensation in combination with the temperature correction unit output value. This design ensures the accuracy of the organic matter thermal evolution parameter inversion at different burial depth layers.

[0051] Specifically, the hydrocarbon source physical property interpretation system based on infrared spectral absorption characteristics provided by the present application, the geological interpretation module is configured to: When the thermal maturity parameter changes by more than a set scale in the vertical depth gradient, activate the fault structure analysis protocol; Receive the hydrogen index parameter transmitted by the physical property mapping module, and when the product of the hydrogen index and the hydrocarbon generation activation energy dispersion coefficient reaches the trigger threshold, start the particle swarm optimization algorithm to perform multi-objective parameter inversion; Wherein, the initial population size of the particle swarm optimization algorithm is constrained by the half-peak width quantification value output by the feature separation module.

[0052] The geological interpretation module receives the thermal maturity parameter and hydrogen index data transmitted by the physical property mapping module. The module monitors the gradient change characteristics of the thermal maturity parameter along the vertical depth, and when the change amplitude exceeds the geological structure anomaly scale, the fault structure analysis protocol is activated. The protocol performs stratigraphic separation calculation and structural stress field simulation to identify the geological fault characteristics of the spatial mutation area of the spectral data.

[0053] The module synchronously receives the hydrogen index parameter and the hydrocarbon generation activation energy dispersion coefficient data. When the product of the two reaches the geological model inversion trigger threshold, the particle swarm optimization algorithm is started to perform multi-objective parameter inversion. The inversion process simultaneously optimizes the kinetic parameters of the hydrocarbon yield curve and the reservoir quality classification coefficient to generate a comprehensive evaluation result of the hydrocarbon generation potential and resource distribution.

[0054] The initial population size of the particle swarm optimization algorithm is dynamically constrained by the half-peak width quantification value transmitted by the feature separation module. The half-peak width quantification value reflects the interference strength of clay minerals, and the strength value is linearly related to the algorithm population size. The mapping establishes a linear relationship. The constraint mechanism ensures that the search space is expanded under strong interference conditions to avoid model distortion caused by local optimal solutions.

[0055] The fault structure analysis protocol and multi-objective inversion form a collaborative decision. The fault location information is used as a boundary condition to inject the particle swarm optimization process, limiting the parameter search range of the hydrocarbon yield curve near the fault zone. At the same time, the optimization result is used to correct the fault stress field model in reverse, updating the spatial distribution of the structural sealing coefficient.

[0056] The half-peak width quantization value continuously adjusts the population characteristics in the inversion cycle. During the algorithm iteration process, the body variation probability and the crossover factor are dynamically adjusted according to the real-time input quantization value. This design makes the optimization process adapt to the fluctuation of mineral interference intensity, and maintains the stability of model convergence.

[0057] The spatial coordinate index runs through the whole process of geological interpretation. The fault analysis result is bound to the three-dimensional coordinate data, and the multi-target inversion output is stored according to the stratigraphic index. The coordinate driving mechanism ensures that the spatial position of the geological model and the spectral collection point is strictly matched.

[0058] Specifically, the hydrocarbon source physical property interpretation system based on infrared spectral absorption characteristics provided by the application is characterized in that the feedback channel of the geological interpretation module and the feature separation module is configured as: Analyzing the carbonate enrichment layer distribution information in the three-dimensional quality grading result; When the carbonate enrichment ratio is higher than the critical value and the silicate template matching degree is lower than the current threshold value: activating the transfer learning mechanism of the dynamic interference database; updating the convolutional neural network weight parameters according to the hydrocarbon yield inversion residual; inputting the updated weight parameters into the spectral peak decoupling unit to perform subsequent decoupling calculation.

[0059] The geological interpretation module analyzes the carbonate enrichment layer distribution data of the three-dimensional quality grading output. The data contains mineral enrichment ratio information and silicate template matching degree index of different geological horizons. When it is detected that the carbonate enrichment ratio of a specific horizon exceeds the critical threshold value and the silicate matching degree is lower than the current discrimination standard, the feedback channel activates the transfer learning mechanism of the dynamic interference database.

[0060] The transfer learning mechanism loads the spectral data of the same horizon of the adjacent exploration well. This operation calls the spectral template and the corresponding mineral combination label in the historical exploration database, and updates the mineral feature template library through the convolutional neural network fine-tuning algorithm. The updated template library enhances the recognition ability of the system to complex mineral paragenetic combination, and eliminates the interference misjudgment caused by local lithological difference.

[0061] The hydrocarbon yield inversion residual data is input into the weight updating calculation unit. This unit calculates the weight correction amount of the fully connected layer of the convolutional neural network according to the deviation amplitude of the inversion model output value and the measured value. The correction algorithm adopts the principle of gradient back propagation, and the greater the residual, the greater the weight adjustment amplitude. The updated weight parameters are packaged as structured data packets and transmitted to the feature separation module in real time.

[0062] The spectral peak decoupling unit of the feature separation module receives the updated weight parameters. The unit loads the new weight file to replace the original neural network model parameters before performing the next cycle of decoupling calculation. The loading process resets the saddle point detection counter synchronously, so that the decoupling process is restarted based on the optimized feature extraction capability.

[0063] The carbonate enrichment layer coordinate data and the weight update mechanism form spatial cooperation. The transfer learning mechanism preferentially calls the spectral data of the surrounding area of the enrichment layer, and the weight correction calculation introduces the spatial weight coefficient of the layer. This design strengthens the focusing optimization capability of the system on the abnormal area of the mineral, and improves the interference recognition accuracy of the target layer.

[0064] The feedback channel establishes a closed-loop iteration timer. After the weight update transmission is completed, the periodic count is automatically started, and when the number of continuous iterations reaches the set value, the model stability evaluation is triggered. The evaluation result determines whether to terminate the feedback process, so as to avoid resource redundancy caused by invalid iteration.

[0065] Specifically, the data transmission channel of the feature separation module and the property mapping module of the hydrocarbon source property interpretation system based on infrared spectral absorption characteristics is configured as: When the decoupled spectral data carries a clay interference label, a verification trigger instruction is sent to the property mapping module; The thermal maturity verification unit of the property mapping module activates the clay correction factor calculation protocol according to the instruction; Wherein, the transmission data contains the half-peak width quantitative value and three-dimensional coordinate information, and the trigger instruction and the clay interference label are generated at the same time.

[0066] The feature separation module generates decoupled spectral data while detecting the state of the clay interference label. When the data packet carries an effective interference label, the transmission channel immediately generates a verification trigger instruction signal. The instruction signal and the decoupled data packet are packaged using the same transmission protocol to ensure the spatiotemporal consistency of the signal and the data.

[0067] The data transmission channel bundles and transmits the decoupled spectral matrix, the half-peak width quantitative value, the three-dimensional coordinate index, and the verification instruction. The packaging structure includes a data header identifier and a verification field, and the header identifier indicates the activation state of the clay interference label. The property mapping module analyzes the instruction priority based on the header identifier to realize the cooperative analysis of data and instructions.

[0068] The thermal maturity verification unit of the property mapping module monitors the instruction signal in real time. When the verification trigger instruction is received, the unit interrupts the current calculation thread and activates the clay correction factor calculation protocol. The protocol calls the three-dimensional coordinate associated geotemperature gradient parameters and the half-peak width quantitative value to start the correction factor iterative operation.

[0069] The synchronization generation mechanism of the interference tag and the verification instruction eliminates the processing delay. The feature separation module creates an instruction timestamp while writing the tag identification, and the physical property mapping module matches the instruction with the corresponding data packet through the timestamp. This design avoids the data alignment error of the traditional polling mechanism and guarantees the immediacy of the correction process.

[0070] The three-dimensional coordinate index drives the layer-specific verification. The verification unit calls the coordinate-bound lithology database and calculates the layer adaptive correction parameter in combination with the half-peak width quantitative value. The spatial constraint mechanism makes the correction factor adapt to the difference in thermodynamic characteristics of different buried depth strata and eliminates the model deviation caused by homogenization processing.

[0071] The transmission channel is built-in feedback confirmation mechanism. After the physical property mapping module completes the correction calculation, it sends a verification completion status code to the feature separation module. The feature separation module updates the life cycle marker of the interference tag according to the status code, realizing cross-module process collaborative management.

[0072] Specifically, the hydrocarbon source property interpretation system based on infrared spectral absorption characteristics disclosed in the present application, the closed loop channel of the geological interpretation module and the feature separation module is configured as: After the particle swarm optimization algorithm outputs the optimal solution of the hydrocarbon yield, the residual change rate of the current decoupling period is calculated; When the residual change rate exceeds the set amplitude, the full connection layer weight parameters of the convolutional neural network are corrected; The updated weight parameters are immediately input into the feature separation module to start the next round of saddle point detection iteration; Wherein, the weight correction amount is negatively correlated with the fitting degree of the hydrocarbon yield curve.

[0073] After the particle swarm optimization algorithm outputs the optimal solution of the hydrocarbon yield, the geological interpretation module calculates the change rate of the spectral inversion residual of the current decoupling period. The change rate is obtained by comparing the mean square error difference between the adjacent period decoupling data and the measured spectrum, and its value reflects the degree of accuracy decay of the feature separation module.

[0074] When the residual change rate exceeds the stability threshold of the geological model, the weight correction unit starts the parameter adjustment of the full connection layer of the convolutional neural network. The correction amount is dynamically calculated according to the fitting degree of the hydrocarbon yield curve, and the higher the fitting degree, the smaller the weight adjustment amplitude, and the two form a negative correlation constraint. This design maintains the stability of the model while avoiding the feature drift caused by excessive correction.

[0075] The updated weight parameters are packaged as structured data packets and transmitted to the feature separation module in real time through the closed loop channel. The transmission protocol includes the weight version identifier and the residual change rate data packet header, and the feature separation module verifies the data validity according to the packet header information and executes parameter loading.

[0076] The spectral peak decoupling unit of the feature separation module resets immediately after receiving the new weight parameters. The unit clears the current saddle point detection buffer area and loads the updated convolutional neural network model. The loading completion signal triggers a new round of saddle point detection iteration, so that the decoupling calculation restarts based on the optimized feature extraction capability.

[0077] The three-dimensional coordinate index runs through the closed-loop correction process. The weight correction unit calls the coordinate-related lithology data, and uses different correction coefficients for different geological horizons. The transmission data packet binds the coordinate index field to ensure that the updated weight and the spatial position are strictly matched.

[0078] The residual change rate monitoring and iteration start form a periodical cooperative management. The system sets an iteration period counter, and activates the model convergence evaluation when the number of continuous corrections reaches the set value. The evaluation result determines whether to terminate the closed-loop process, so as to maintain the dynamic balance of system computing resources.

[0079] In the second aspect, referring to Figure 1 The present application provides a hydrocarbon source property interpretation method based on infrared spectrum absorption characteristics, which is applied to the hydrocarbon source property interpretation method based on infrared spectrum absorption characteristics, and includes the following steps: Step 1, collecting in-situ gas infrared spectrum and core micro-area infrared spectrum of the hydrocarbon source rock, generating a 1300-3500 nm waveband time spectrum, and obtaining a transmission mode spectrum matrix and a reflection mode spectrum matrix of a 50 mu m micro-area; Step 2, identifying mineral interference and decoupling organic characteristics: detecting 1700- The second-order derivative spectrum saddle point interval, when the number of saddle points is greater than or equal to 3, the convolutional neural network decoupling overlapping peaks are started, and when the 3400 Waveband half-peak width is greater than 120 When the clay interference label is activated, the half-peak width quantization value is recorded; Step 3, performing dynamic correction of physical property parameters: receiving the separated features with the clay interference label and the half-peak width quantization value, when the aromatic ring absorption peak symmetry index is less than 0.85 and the difference between the fat chain peak intensity is greater than 20%, the clay correction factor is triggered; Step 4, constructing a geological resource grading model: when the hydrocarbon activation energy dispersion coefficient is greater than 0.5, the particle swarm optimization algorithm is started to invert the hydrocarbon yield curve, and the silicate template matching threshold is updated to 85% based on the mineral abnormal area distribution data; Step 5, correcting the convolutional neural network weight parameters according to the optimal solution of the hydrocarbon yield, inputting the updated weight parameters into the next period decoupling calculation, and returning to step 2 for iteration.

[0080] The present application solves the problem of deviation in organic matter parameter inference caused by the superposition of inorganic mineral absorption characteristics through a triple technical mechanism: Mineral interference feature dynamic decoupling mechanism. The characteristic separation module constructs an inorganic mineral waveband feature template library, and controls the convolutional neural network to separate the overlapping absorption peaks through the second derivative spectrum saddle point threshold. When the mineral template matching degree is insufficient, the transfer learning mechanism loads the adjacent area data for supplementary training, so that the system adapts to the regional mineral combination changes. The decoupling process synchronously detects the half-peak width anomaly of the hydroxyl stretching vibration waveband, and generates a clay interference label with a quantitative value. This mechanism accurately removes the absorption interference of silicates, carbonates and other minerals in specific wavebands, ensuring the complete extraction of organic functional group characteristic peaks.

[0081] Layer-specific dynamic compensation mechanism. After receiving the separated feature data with three-dimensional coordinates, the physical property mapping module performs dual-factor verification of the symmetry of aromatic ring absorption peaks and the peak intensity ratio of aliphatic chains. When both factors exceed the limit, the clay correction factor is activated, and the compensation strength is dynamically adjusted based on the half-peak width quantitative value. The spatial coordinate index drives the call of the layer temperature gradient parameter, realizing the depth-adaptive correction of total organic carbon content and hydrogen index. The synergistic effect of mineral interference quantitative indicators and spatial parameters eliminates the stratigraphic adaptability defects of the uniformization compensation model.

[0082] Closed-loop feedback optimization mechanism. The geological interpretation module updates the convolutional neural network weights based on the hydrocarbon yield inversion residual, and improves the feature separation accuracy through cross-cycle iteration. The three-dimensional quality grading results analyze the distribution of carbonate-rich layers, triggering dynamic updating of the mineral template library; the particle swarm optimization algorithm outputs the optimal solution to constrain the weight correction amount, forming a negative feedback balance between decoupling accuracy and model fitting degree. This closed-loop system continuously reduces the parameter transmission error caused by mineral interference, achieving autonomous evolution of the accuracy of organic matter abundance and type inference.

[0083] The above technical chain breaks through the limitations of traditional static spectral analysis: dynamic synergy of mineral interference recognition and feature decoupling eliminates absorption peak positioning deviation; layer-constrained compensation mechanism suppresses parameter quantitative distortion; geological model feedback-driven closed-loop optimization continuously improves the system's anti-interference ability. Ultimately, it realizes the collaborative optimization of organic matter qualitative and quantitative parameter inference.

Claims

1. A hydrocarbon source property interpretation system based on infrared spectral absorption characteristics, characterized in that, It includes a spectral data acquisition module, a feature separation module, a physical property mapping module, and a geological interpretation module; The spectral data acquisition module is configured to acquire in-situ gas infrared spectra and core micro-area infrared spectra of source rock samples. The feature separation module is configured to be configured as follows: Test 1700- Saddle points of the second derivative spectrum of the interval; When the second derivative spectrum has saddle points ≥ 3, the convolutional neural network is activated to decouple the overlapping absorption peaks. At 3400 Band half-width > 120 Activate the clay interference tag at the time; Output the separation characteristics of organic functional groups with clay interference tags; The property mapping module is configured to be configured as follows: Receive separation characteristics with clay interference tags; When the difference between the symmetry of the aromatic ring absorption peak and the peak intensity ratio of the aliphatic chain is greater than 20%, the clay correction factor is triggered. Output the corrected total organic carbon content, thermal maturity, and hydrogen index parameters; The geological interpretation module is configured to be configured as follows: When the dispersion coefficient of the hydrocarbon generation activation energy is greater than 0.5, the particle swarm optimization algorithm is activated to invert the hydrocarbon yield curve. Receive mineral anomaly zone distribution data and update the silicate template matching threshold to 85%; The optimal solution for hydrocarbon yield output by the particle swarm optimization algorithm is used to correct the weight parameters of the convolutional neural network. The updated weight parameters are then input into the feature separation module for the next cycle of decoupling calculation.

2. The hydrocarbon source property interpretation system based on infrared spectral absorption characteristics according to claim 1, characterized in that, The spectral data acquisition module includes: The in-situ gas acquisition unit separates hydrocarbon gases at the drilling fluid outlet through a semi-permeable membrane separator. After the carbonate dust is removed by molecular sieve filtration, the gas is input into the gas absorption cell to generate a time-series spectrum in the 1300-3500nm band. The rock sample acquisition unit uses argon ion polishing to polish the rock core sample and scans it on a micro-infrared platform to output a dual-mode transmission and reflection spectral matrix of a 50μm×50μm micro-area; the feature separation module receives the time-series spectrum and the dual-mode spectral matrix and establishes a three-dimensional mapping relationship according to the well depth coordinates.

3. The hydrocarbon source property interpretation system based on infrared spectral absorption characteristics according to claim 2, characterized in that, The feature separation module includes: Dynamic interference database, storing silicate 1100-1000 and carbonates 1500-1400 Band feature templates are used, and the transfer learning mechanism is activated when the band feature template matching degree is less than 90%. Spectral peak decoupling unit, calculation 1700- Saddle points of the second derivative spectrum in the interval: When the saddle point of the second derivative spectrum is ≥3, the residual connection convolutional neural network is activated, and the input data is the parallel data of the original spectrum and the second derivative spectrum. Output the organic functional group separation features with three-dimensional coordinates and clay interference labels to the physical property mapping module.

4. The hydrocarbon source property interpretation system based on infrared spectral absorption characteristics according to claim 3, characterized in that, The peak decoupling unit is configured as follows: At 3400 Band half-width > 120 The clay interference tag is activated at the specified time, and the half-peak width quantization value is recorded. The quantized value and the decoupled spectral data with three-dimensional coordinates are synchronously transmitted to the property mapping module; Among them, the total organic carbon compensation mechanism of the clay interference tag directly associated with the physical property mapping module has a half-width quantization value > 130. The compensation coefficient is increased to 1.2 times.

5. The hydrocarbon source property interpretation system based on infrared spectral absorption characteristics according to claim 4, characterized in that, The property mapping module is configured as follows: Receive decoupled spectral data carrying three-dimensional coordinates and full width at half maximum (FWHM) quantization values; When the symmetry index of the aromatic ring absorption peak is <0.85 and the difference in peak intensity ratio of the aliphatic chain is >20%, the clay correction factor is activated. The alkyl / carboxyl peak area ratio was converted into the hydrogen index using a pre-trained regression model. The half-width quantization value dynamically adjusts the temperature correction parameter of the regression model; for every 10 units increase in the half-width quantization value... The temperature correction slope was increased by 0.

15.

6. The hydrocarbon source property interpretation system based on infrared spectral absorption characteristics according to claim 5, characterized in that, The geological interpretation module is configured as follows: When the change in the thermal maturity parameter in the vertical depth gradient exceeds the set scale, the fault structure analysis protocol is activated. The system receives the hydrogen index parameter transmitted by the property mapping module. When the product of the hydrogen index and the dispersion coefficient of the hydrocarbon generation activation energy reaches the trigger threshold, the particle swarm optimization algorithm is started to perform multi-objective parameter inversion. The initial population size of the particle swarm optimization algorithm is constrained by the half-peak width quantization value output by the feature separation module.

7. The hydrocarbon source property interpretation system based on infrared spectral absorption characteristics according to claim 6, characterized in that, The feedback channels of the geological interpretation module and the feature separation module are configured as follows: Analysis of carbonate enrichment layer distribution information in three-dimensional quality grading results; When the carbonate enrichment ratio is higher than the critical value and the silicate template matching degree is lower than the current threshold: Activate the transfer learning mechanism of the dynamic interference database; The weight parameters of the convolutional neural network are updated based on the hydrocarbon yield inversion residuals; The updated weight parameters are input into the spectral peak decoupling unit to perform subsequent decoupling calculations.

8. The hydrocarbon source property interpretation system based on infrared spectral absorption characteristics according to claim 7, characterized in that, The data transmission channel between the feature separation module and the property mapping module is configured as follows: When the decoupled spectral data carries a clay interference tag, a verification trigger command is sent to the property mapping module. The thermal maturity verification unit of the physical property mapping module activates the clay correction factor calculation protocol according to the instruction. The transmitted data includes the half-peak width quantization value and three-dimensional coordinate information, and the trigger command and clay interference label are generated simultaneously.

9. The hydrocarbon source property interpretation system based on infrared spectral absorption characteristics according to claim 8, characterized in that, The closed-loop channel between the geological interpretation module and the feature separation module is configured as follows: After the particle swarm optimization algorithm outputs the optimal solution for hydrocarbon yield, the rate of change of residual in the current decoupling cycle is calculated. When the rate of change of the residual exceeds the set range, the weight parameters of the fully connected layer of the convolutional neural network are adjusted. The updated weight parameters are immediately input into the feature separation module to start the next round of saddle point detection iteration; Among them, the weight correction amount is negatively correlated with the goodness of fit of the hydrocarbon yield curve.

10. A method for interpreting hydrocarbon source properties based on infrared spectral absorption characteristics, applied to the method for interpreting hydrocarbon source properties based on infrared spectral absorption characteristics as described in any one of claims 1 to 9, characterized in that, include: Step 1: Collect in-situ gas infrared spectra and core micro-area infrared spectra of source rocks to generate time-series spectra in the 1300-3500nm band, and obtain the transmission mode spectral matrix and reflection mode spectral matrix of 50μm micro-area. Step 2, Identify mineral interference and decouple organic features: Detect 1700- The second derivative spectrum saddle points in the interval are used to initiate a convolutional neural network to decouple overlapping peaks when the number of saddle points is ≥3, and at 3400 Band half-width > 120 The clay interference label is activated at time, and the full width at half maximum (FWHM) quantization value is recorded. Step 3, perform dynamic correction of physical property parameters: receive the separation characteristics with clay interference tags and half-peak width quantization values, and trigger the clay correction factor when the symmetry index of the aromatic ring absorption peak is <0.85 and the difference in peak intensity ratio of the aliphatic chain is >20%. Step 4, construct a geological resource classification model: when the dispersion coefficient of hydrocarbon generation activation energy is >0.5, start the particle swarm optimization algorithm to invert the hydrocarbon yield curve, and update the silicate template matching threshold to 85% based on the mineral anomaly zone distribution data; Step 5: Correct the weight parameters of the convolutional neural network based on the optimal solution for hydrocarbon yield, input the updated weight parameters into the next cycle of decoupling calculation, and return to step 2 to perform iteration.

Images