A method and system for on-line near-infrared spectrum detection of biomass raw material components

Abstract

The application provides an online near-infrared spectrum detection method and system for biomass raw material components, and relates to the technical field of component detection. Firstly, the particle size distribution and the near-infrared spectrum data of the biomass raw material are acquired in real time, and the two types of data are time-stamped and aligned to establish a correlation mapping. Secondly, based on the mapping data, the channel spacing of a multi-channel adjustable distance optical attenuator is dynamically adjusted to generate standardized spectrum data that eliminates particle size interference. Then, by assigning differentiated weights to the characteristic wavebands of each main component, enhanced characteristic spectrum data is constructed. Finally, a multi-output regression model is established using pre-set samples, and the contents of cellulose, hemicellulose, lignin and other components are simultaneously inverted by inputting the enhanced characteristic spectrum data into the model, thereby realizing rapid and accurate online detection of the components of the biomass raw material in the continuous conveying process.

Description

Technical Field

[0001] This application relates to the field of component detection technology, and in particular to a near-infrared spectroscopy online detection method and system for components of biomass raw materials. Background Technology

[0002] In fields such as biomass energy, biochemicals, and feed processing, the stability of raw material composition directly affects the efficiency of subsequent processes and product quality. When biomass raw materials are continuously transported from storage silos to processing equipment such as crushing, pyrolysis, or fermentation equipment, their chemical composition often fluctuates due to diverse sources, seasonal variations, or differences in pretreatment. To achieve automated and intelligent process control, there is an urgent need for a technology capable of acquiring real-time, non-destructive, and continuous information on the multi-component content of raw materials during transportation. This would support dynamic proportioning adjustments, process parameter optimization, and quality traceability, placing high demands on the response speed, environmental adaptability, and simultaneous multi-component analysis capabilities of the detection methods.

[0003] To address this need, existing solutions include online detection systems based on near-infrared spectroscopy combined with fixed optical path structures. These systems typically integrate transmission or reflection near-infrared probes above the conveyor pipe or belt, acquiring the overall spectral signal during material flow. The content of each component is then directly inverted from the raw spectrum using a pre-trained partial least squares regression (PLS) or multilayer neural network model. This method eliminates the lag between offline sampling and laboratory analysis, achieving true online monitoring, and has already been deployed in some biomass processing lines.

[0004] The existing scheme has significant drawbacks: First, due to the common problems of uneven particle size, fluctuating bulk density, and differences in surface roughness in biomass raw materials, the collected near-infrared spectra are easily affected by scattering effects and changes in optical path, resulting in the same component exhibiting significantly different spectral responses under different physical states. Second, the fixed optical path structure cannot dynamically compensate for optical signal distortions caused by the physical state of the material, thus limiting the model's generalization ability. In particular, the prediction accuracy drops sharply when raw material batches change or humidity changes abruptly, making it difficult to meet the stringent requirements for detection reliability in highly stable industrial processes. Summary of the Invention

[0005] The purpose of this application is to provide a near-infrared spectroscopy online detection method and system for biomass raw material components, so as to solve the problem that the existing technology is difficult to meet the stringent requirements for detection reliability in highly stable industrial processes.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a near-infrared spectroscopy online detection method for biomass raw material components, comprising:

[0007] Obtain particle size distribution data and near-infrared spectral data of the pre-placed biomass raw materials;

[0008] The particle size distribution data and the near-infrared spectral data are time-stamp aligned to obtain the correlation mapping data between the particle size distribution data and the near-infrared spectral data;

[0009] The spacing between each channel in the preset multi-channel adjustable optical attenuator is adjusted using the aforementioned correlation mapping data to obtain standardized spectral data;

[0010] Differential weights corresponding to the characteristic bands of each major component of the pre-set biomass raw material are assigned to the standardized spectral data to generate enhanced characteristic spectral data;

[0011] A multi-output regression model was established using spectral data of biomass samples with known component contents.

[0012] The enhanced characteristic spectral data is input into the multi-output regression model to generate content data of multiple components in the pre-set biomass raw material.

[0013] Optionally, adjusting the spacing of each channel in a preset multi-channel adjustable-distance optical attenuator using the associated mapping data to obtain standardized spectral data includes:

[0014] The correlation mapping data is analyzed to determine the particle size distribution interval with the highest frequency of particle size values ​​in the correlation mapping data as the first particle size distribution interval;

[0015] The first particle size distribution range is identified to obtain the dominant range identifier;

[0016] The target spacing value is obtained by accessing the preset interval spacing mapping table based on the dominant interval identifier.

[0017] The target spacing value is encoded to generate a pulse width modulation signal;

[0018] The pulse width modulation signal is loaded into the actuator of a preset multi-channel adjustable optical attenuator to generate near-infrared spectral data as standardized spectral data.

[0019] Optionally, the step of analyzing the correlation mapping data to determine the particle size distribution interval with the highest frequency of particle size values ​​in the correlation mapping data as the first particle size distribution interval includes:

[0020] The preset statistical time window is divided into multiple consecutive time slices to determine the time slice sequence of the current statistical period;

[0021] Traverse each time slice in the time slice sequence and extract the granularity value of each associated mapping data in each time slice to generate a set of granularity values ​​for each time slice.

[0022] The set of particle size values ​​and the preset particle size intervals to which each set of particle size values ​​belongs are matched, and a cumulative quantity corresponding to each preset particle size interval is generated based on the matching result.

[0023] The ratio between the cumulative quantity of each particle size and the total number of particle size values ​​in each set of particle size values ​​is calculated, and the ratio calculation result is used as the distribution density of each preset particle size interval.

[0024] Calculate the confidence coefficient of the distribution characteristics of each particle size value in the current statistical period based on the distribution density;

[0025] Dynamic weighting is applied to the cumulative quantities based on the confidence coefficients to generate the particle size distribution interval with the highest frequency of particle size values ​​as the first particle size distribution interval.

[0026] Optionally, the step of dynamically weighting each of the cumulative quantities according to each of the confidence coefficients to generate the particle size distribution interval with the highest frequency of the particle size values ​​as the first particle size distribution interval includes:

[0027] Each confidence coefficient is input into a preset confidence weight mapping relationship to generate a dynamic weight factor corresponding to each preset particle size interval to which each set of particle size values ​​belongs.

[0028] The cumulative quantity is multiplied by the dynamic weighting factor corresponding to each cumulative quantity to obtain the weighted cumulative quantity of each preset particle size interval;

[0029] The numerical values ​​of each weighted cumulative quantity are compared, and the preset particle size interval corresponding to the weighted cumulative quantity with the largest value is taken as the particle size distribution interval with the highest frequency of particle size value occurrence, and is taken as the first particle size distribution interval.

[0030] Optionally, encoding the target spacing value to generate a pulse width modulation signal includes:

[0031] The target spacing value and the preset reference spacing value are used to perform a difference calculation, and the result of the difference calculation is used as the spacing difference.

[0032] The spacing difference is input into a preset quantization level correspondence to generate a duty cycle level corresponding to the spacing difference;

[0033] The signal conduction time percentage corresponding to the duty cycle level is determined from the preset correspondence between duty cycle level and signal conduction time percentage.

[0034] The preset pulse period timing unit is filled according to the signal conduction time ratio to generate a pulse width modulation signal.

[0035] Optionally, assigning differentiated weights to the characteristic bands corresponding to each major component of the pre-set biomass raw material to the standardized spectral data to generate enhanced characteristic spectral data includes:

[0036] The standardized spectral data is divided into multiple spectral data segments based on the wavelength of the standardized spectral data.

[0037] Each of the spectral data segments is matched with the characteristic bands corresponding to each major component of the pre-set biomass raw material, and the spectral data segments belonging to each of the characteristic bands are identified from the matching results.

[0038] The weighted spectral data segments are obtained by multiplying each spectral data segment with the weight value corresponding to each spectral data segment from the preset set of weight values.

[0039] The spectral data segments that do not belong to each of the aforementioned characteristic bands are multiplied by a preset reference weight value to obtain a reference spectral data segment.

[0040] The weighted spectral data segments and the reference spectral data segments are recombined according to the wavelength order of the weighted spectral data segments and the wavelength order of the reference spectral data segments to generate enhanced feature spectral data.

[0041] Optionally, inputting the enhanced characteristic spectral data into the multi-output regression model to generate content data of multiple components in the pre-set biomass raw material includes:

[0042] Extract wavelength intensity values ​​from the enhanced spectral data;

[0043] The wavelength intensity values ​​are arranged and combined according to the wavelength order to generate a spectral feature vector;

[0044] The initial component content vector is generated by multiplying the spectral feature vector and the parameter matrix in the multi-output regression model.

[0045] The initial component content vector and the bias vector in the multi-output regression model are added together to generate the intermediate component content vector.

[0046] The intermediate component content vector is nonlinearly transformed to generate the final component content vector.

[0047] Each element value in the final component content vector is converted into the content data of each component in the pre-set biomass raw material corresponding to each element value.

[0048] Secondly, this application provides an online near-infrared spectroscopy detection system for biomass raw material components, comprising:

[0049] The acquisition module is used to acquire particle size distribution data and near-infrared spectral data of the pre-set biomass raw materials;

[0050] The processing module is used to perform time stamp alignment processing on the particle size distribution data and the near-infrared spectral data to obtain the correlation mapping data between the particle size distribution data and the near-infrared spectral data;

[0051] The adjustment module uses the associated mapping data to adjust the spacing of each channel in the preset multi-channel adjustable optical attenuator to obtain standardized spectral data.

[0052] The allocation module assigns differentiated weights to the characteristic bands corresponding to the main components of the pre-set biomass raw materials to the standardized spectral data in order to generate enhanced characteristic spectral data.

[0053] A module is established to build a multi-output regression model using spectral data of biomass samples with known component contents;

[0054] The generation module inputs the enhanced characteristic spectral data into the multi-output regression model to generate the content data of multiple components in the preset biomass raw materials.

[0055] Thirdly, this application provides an electronic device, comprising:

[0056] Memory, used to store computer programs;

[0057] A processor is configured to execute the computer program to implement the steps of an online near-infrared spectroscopy detection method for biomass raw material components as described in the first aspect above.

[0058] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the near-infrared spectroscopy online detection method for biomass raw material components as described in the first aspect above.

[0059] This application provides an online near-infrared spectroscopy detection method for biomass raw material components. It acquires the particle size distribution and near-infrared spectral data of the biomass raw material, aligns them with timestamps, and establishes a correlation mapping between the two types of data. Based on this mapping data, it dynamically adjusts the channel spacing of a multi-channel adjustable-spacing optical attenuator to achieve standardized processing of spectral signals from raw materials with different particle sizes. Furthermore, it generates enhanced characteristic spectral data by assigning differentiated weights to the characteristic bands of each major component to the standardized spectral data. Finally, it uses a multi-output regression model established with preset samples to simultaneously invert the content of multiple components. This method effectively solves the problems of inaccurate spectral signals and low efficiency in single-component detection caused by dynamic changes in particle size distribution during continuous transport of biomass raw materials.

[0060] Furthermore, by analyzing the correlation mapping data, the most frequently occurring particle size distribution interval is determined and a dominant interval identifier is generated. Based on this identifier, the interval spacing mapping table is accessed to obtain the target spacing value. After encoding, a pulse width modulation signal is generated to drive the optical attenuator actuator, thereby realizing the automatic adjustment of optical system parameters according to the particle size distribution characteristics of the raw material. This ensures that the spectral signal intensity corresponding to raw materials of different particle sizes is stabilized within the optimal response range of the detection equipment, providing a standardized data foundation for subsequent spectral data processing that is not affected by particle size changes. Attached Figure Description

[0061] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A schematic flowchart of an online near-infrared spectroscopy detection method for biomass raw material components provided in this application embodiment;

[0063] Figure 2 A schematic flowchart of another near-infrared spectroscopy online detection method for biomass raw material components provided in this application embodiment;

[0064] Figure 3 This is a schematic diagram of the structure of an online near-infrared spectroscopy detection system for biomass raw material components provided in an embodiment of this application. Detailed Implementation

[0065] In industrial monitoring scenarios involving continuous biomass feedstock transportation, the fundamental challenge facing existing near-infrared spectroscopy technology stems from the dynamic changes in the physical properties of the materials. Biomass feedstocks exhibit complex and fluctuating particle size distributions during transportation. These physical variations directly lead to unpredictable scattering and attenuation of near-infrared light, resulting in raw spectral signals containing a large amount of interference information unrelated to the composition. Traditional methods typically employ fixed optical configurations and single spectral processing algorithms, failing to compensate for optical path differences caused by particle size variations at the physical level, and struggling to effectively separate particle size interference from component characteristic signals during data processing. Furthermore, existing technologies often employ sequential detection of each component, which is not only inefficient but also ignores the interrelationships between components, making it difficult to meet the real-time and accuracy requirements of online monitoring.

[0066] To address these issues, the research and development approach of this invention is as follows: First, a real-time correlation mapping between particle size distribution and spectral data is established. This mapping data is then used to dynamically control the physical parameters of a multi-channel adjustable-distance optical attenuator, actively compensating for optical signal distortion caused by particle size differences along the light source propagation path, and generating standardized spectral data. Next, the characteristic spectral bands of the main components of biomass are selectively enhanced to highlight effective information and suppress irrelevant noise. Finally, the enhanced spectral features are synchronously analyzed using a multi-output regression model, outputting the accurate content of multiple components at once. This forms a complete technical solution from physical compensation to data processing, from feature enhancement to collaborative inversion, enabling rapid and accurate detection of multiple components under dynamic operating conditions.

[0067] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0068] The core of this application is to provide an online near-infrared spectroscopy method for detecting components in biomass raw materials, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0069] S101. Obtain particle size distribution data and near-infrared spectral data of the pre-placed biomass raw materials.

[0070] Among them, particle size distribution data refers to the data obtained by real-time measurement of biomass raw material particle size using a laser particle size analyzer, including statistical information reflecting the number or volume percentage of particles in different particle size ranges, which is used to analyze the uniformity and dispersion state of raw materials based on particle physical morphology characteristics; near-infrared spectral data refers to the data obtained by collecting the absorption or reflection response of biomass raw materials to near-infrared light using a near-infrared spectrometer, including the sequence of light intensity values ​​at different wavelengths, which is used to identify the chemical information of components such as cellulose, hemicellulose and lignin in raw materials based on molecular vibration characteristics.

[0071] In this embodiment, a laser particle size analyzer installed upstream of the biomass raw material conveyor belt is first used to scan and detect the continuously flowing raw material in real time. The instrument emits a laser beam and receives the diffraction signal generated by the raw material particles. By analyzing the correspondence between the diffraction angle and the particle size, the volume distribution percentage of each particle size range is calculated, thereby obtaining particle size distribution data that reflects the physical morphological characteristics of the raw material.

[0072] Secondly, a near-infrared spectroscopy detection device deployed downstream of the conveyor belt is used. Its built-in broadband light source illuminates the surface of the raw material, and an array detector receives the light signal absorbed and scattered by the raw material. After the light is processed by the spectrometer and photoelectric conversion, a sequence of light intensity values ​​at different near-infrared wavelengths is obtained, forming near-infrared spectral data containing information on the chemical composition of the raw material.

[0073] Finally, the data collected by the laser particle size analyzer and the spectral detection device are transmitted to the central controller in real time via an industrial fieldbus. At the data receiving end, a timestamp accurate to the millisecond level is added to each set of particle size distribution data and near-infrared spectral data to ensure that the two types of data are completely corresponding in the time dimension, forming a raw dataset with spatiotemporal consistency, which provides a foundation for establishing data association in the future.

[0074] Specifically, in the raw material quality inspection system of a biomass power plant, firstly, an online laser particle size analyzer installed at the discharge port of the crusher scans the wood chips flowing on the conveyor belt in real time, obtaining a distribution statistical table containing five particle size ranges; secondly, a near-infrared spectral probe installed on the support in the middle of the conveyor belt scans the surface of the passing raw material three times per second across the entire spectrum, collecting absorbance data at 1,200 wavelength points; finally, the data from the particle size analyzer and the spectrometer are transmitted in real time to the central control room server via industrial Ethernet, where the data acquisition program adds millisecond-level timestamps to each set of data, forming a time-stamped raw material characteristic dataset.

[0075] S102. Perform time-stamp alignment processing on the particle size distribution data and the near-infrared spectral data to obtain the correlation mapping data between the particle size distribution data and the near-infrared spectral data.

[0076] Among them, the correlation mapping data refers to the data set obtained by fusing time-domain aligned particle size distribution data and near-infrared spectral data. It includes matching data pairs that reflect the correspondence between the physical and chemical properties of raw materials at the same time point, and is used to establish a quantitative correlation model between particle size distribution range and spectral response characteristics.

[0077] In this embodiment, the system first receives detection data synchronously uploaded from the laser particle size analyzer and the near-infrared spectrometer through the data acquisition system. These data packets all carry timestamp information accurate to the millisecond level. The system automatically parses the data packet structure and extracts the time stamp sequence.

[0078] Secondly, a sliding time window matching algorithm is adopted. Based on a preset time tolerance value, a dynamic matching window is established on a continuous time axis. Particle size distribution data points and near-infrared spectral data points falling within the same time window are precisely paired to generate a data matching table containing time correspondence.

[0079] Finally, the successfully paired particle size distribution data and near-infrared spectral data are structurally integrated through the data fusion engine. The distribution data of each particle size interval is associated and stored with its corresponding spectral feature data according to the time series, forming a spatiotemporally consistent association mapping database.

[0080] Specifically, in the raw material pretreatment workshop of a biomass refinery, the central control system first receives real-time detection data from laser particle size analyzers and near-infrared spectrometers, all of which are time-stamped to the millisecond level. Then, a sliding time window algorithm is used to automatically match particle size distribution data points and near-infrared spectral data points with the same time period marker, based on a set time tolerance. Finally, the successfully matched particle size distribution data and the corresponding spectral feature data are combined and stored to form a complete correlation mapping database containing time, particle size, and spectral dimensions.

[0081] The overall scheme of S102 described above establishes an accurate correspondence between the two different types of data by performing precise timestamp alignment processing on particle size distribution data and near-infrared spectral data. This effectively solves the analysis error problem caused by asynchronous data acquisition time during dynamic detection and provides a reliable data foundation for subsequent spectral standardization processing.

[0082] S103. Apply the associated mapping data to adjust the spacing of each channel in the preset multi-channel adjustable optical attenuator to obtain standardized spectral data.

[0083] Optionally, such as Figure 2 As shown, S103 may specifically include the following steps:

[0084] S1031. Analyze the correlation mapping data and determine the particle size distribution interval with the highest frequency of particle size values ​​in the correlation mapping data as the first particle size distribution interval.

[0085] Specifically, step S1031 includes the following process:

[0086] A preset statistical time window is divided into multiple consecutive time slices to determine the time slice sequence of the current statistical period. Each time slice in the time slice sequence is traversed, and the particle size value of each associated mapping data within each time slice is extracted to generate a set of particle size values ​​for each time slice. Each set of particle size values ​​is matched with each preset particle size interval to which it belongs, and a cumulative quantity corresponding to each preset particle size interval is generated based on the matching result. The ratio of each cumulative quantity to the total number of particle size values ​​in each set of particle size values ​​is calculated, and the ratio result is used as the distribution density of each preset particle size interval. A confidence coefficient for the distribution characteristics of each particle size value within the current statistical period is calculated based on the distribution density. Dynamic weight allocation is performed on each cumulative quantity based on the confidence coefficient to generate the particle size distribution interval with the highest frequency of particle size value occurrence as the first particle size distribution interval.

[0087] The step of dynamically weighting the cumulative quantities according to the confidence coefficients to generate the particle size distribution interval with the highest frequency of particle size values ​​as the first particle size distribution interval specifically includes the following process:

[0088] Each confidence coefficient is input into a preset confidence weight mapping relationship to generate a dynamic weight factor corresponding to each preset particle size interval to which each set of particle size values ​​belongs; each cumulative quantity is multiplied by the dynamic weight factor corresponding to each cumulative quantity to obtain the weighted cumulative quantity of each preset particle size interval; the values ​​of each weighted cumulative quantity are compared, and the preset particle size interval corresponding to the weighted cumulative quantity with the largest value is taken as the particle size distribution interval with the highest frequency of particle size values ​​as the first particle size distribution interval.

[0089] The first particle size distribution interval refers to the particle size range with the highest occurrence frequency determined from the associated mapping data through frequency statistics, including particle size interval data with the highest statistical frequency, which is used to guide the adjustment of optical system parameters based on the main particle size characteristics.

[0090] S1032. The first particle size distribution range is marked to obtain the dominant range mark.

[0091] Among them, the dominant interval identifier refers to the unique identification mark generated for the first particle size distribution interval, including metadata such as interval number, boundary value and statistical characteristics, which is used for fast retrieval and data association in the interval spacing mapping table.

[0092] S1033. Access the preset interval spacing mapping table based on the dominant interval identifier to obtain the target spacing value.

[0093] The target spacing value refers to the optical attenuator spacing setting value obtained by querying the interval spacing mapping table, including the mechanical adjustment parameters determined according to the dominant interval identifier, which are used to guide the precise position control of the optical attenuator.

[0094] S1034. Encode the target spacing value to generate a pulse width modulation signal.

[0095] Specifically, step S1034 includes the following process:

[0096] The target spacing value and the preset reference spacing value are subjected to a difference calculation to obtain the spacing difference value. The spacing difference value is input into a preset quantization level correspondence to generate a duty cycle level corresponding to the spacing difference value. The signal conduction time ratio corresponding to the duty cycle level is determined from the preset duty cycle level and signal conduction time ratio correspondence. The preset pulse period timing unit is filled according to the signal conduction time ratio to generate a pulse width modulation signal.

[0097] Among them, the pulse width modulation signal refers to the electronic control signal generated according to the target spacing value, including square wave signals with specific pulse width characteristics, which are used to drive the precise action of the optical attenuator actuator.

[0098] S1035. The pulse width modulation signal is loaded into the actuator of the preset multi-channel adjustable optical attenuator to generate near-infrared spectral data as standardized spectral data.

[0099] Standardized spectral data refers to near-infrared spectral data acquired after adjusting the optical attenuator, including spectral signals that have undergone optical path standardization, which is used to eliminate the influence of particle size differences on spectral intensity.

[0100] In this embodiment of the application, the following process is first performed in step S1031 to determine the first particle size distribution range:

[0101] The first step is to divide the time-slice sequence. The preset statistical time window is divided into multiple consecutive time slices of equal length, thereby determining the time-slice sequence of the current statistical period. For example, if the preset statistical time window is the most recent 30 seconds, it is divided into time slices of 1 second in length.

[0102] The second step involves extracting and generating a set of particle size values, iterating through each time slice in the time slice sequence. For each time slice, the particle size values ​​contained in all associated mapping data within it are extracted, i.e., the specific particle size value corresponding to each data point. These values ​​are then aggregated to generate a set of particle size values ​​specific to that time slice. For example, within a certain second of a time slice, 150 particle size values ​​may be extracted.

[0103] The third step involves matching and accumulating data. Multiple particle size ranges are preset, such as 0-0.5mm, 0.5-1.0mm, and 1.0-2.0mm. Each value in the particle size data set for each time slice is matched against these preset ranges, and the number of values ​​matching within each range is counted. The results are accumulated across all time slices, generating a cumulative count corresponding to each preset particle size range. For example, the 0.5-1.0mm range was found to match a total of 5000 values ​​across all time slices.

[0104] The fourth step is to calculate the distribution density. Divide the cumulative number of particles in each size range by the total number of particle size values ​​in all time slices within the current statistical period to obtain the distribution density of that range. For example, if the total number is 10,000, then the distribution density of the 0.5-1.0 mm range is 5000 / 10000=0.5, or 50%.

[0105] The fifth step is to calculate the confidence coefficient. Based on the calculated distribution density of each interval, the stability of the particle size distribution characteristics within the current statistical period is assessed. For example, if the distribution density of a certain interval fluctuates very little across all time slices, it is assigned a higher confidence coefficient, such as close to 1.0; if the fluctuation is drastic, it is assigned a lower confidence coefficient, such as close to 0.5. This coefficient can be calculated using a preset calculation model, such as based on variance or entropy.

[0106] The sixth step involves dynamically assigning weights and determining the first interval. This includes inputting the confidence coefficients of each particle size interval into a preset confidence weight mapping relationship to generate corresponding dynamic weight factors. In practice, the preset confidence weight mapping relationship can be a function or lookup table that maps confidence to weight factors. Then, the cumulative quantity of each interval is multiplied by its dynamic weight factor to obtain a weighted cumulative quantity. Finally, the weighted cumulative quantities of all intervals are compared, and the interval with the largest value is determined as the particle size distribution interval with the highest frequency of particle size values, i.e., the first particle size distribution interval. For example, the 0.5-1.0 mm interval has the largest weighted cumulative quantity and is therefore determined as the first particle size distribution interval.

[0107] Secondly, a unique numerical identifier is generated by determining the first particle size distribution interval through step S1032. This identifier is then stored together with metadata such as interval boundary values ​​and frequency of occurrence to form the dominant interval identifier. For example, the system assigns the identifier code "DOM001" to the first particle size distribution interval of 0.5-1.0mm and records its boundary values ​​of 0.5 and 1.0 and its frequency of occurrence of 45% to construct a complete dominant interval identifier.

[0108] Subsequently, the system accesses the preset interval spacing mapping table through the query interface in step S1033, then inputs the dominant interval identifier as the retrieval key into the query system, and finally extracts the corresponding spacing value from the mapping table as the target spacing value. For example, the system queries the mapping table with "DOM001" as the key value and returns the corresponding target spacing value of 3.5mm, completing the data query process.

[0109] Next, in step S1034, the target spacing value is encoded to generate a pulse width modulation (PWM) signal, specifically including:

[0110] The first step is to calculate the spacing difference. This involves calculating the difference between the target spacing value and a preset baseline spacing value. For example, if the target spacing value is 3.5mm and the preset baseline spacing value is 2.0mm, the spacing difference = 3.5mm - 2.0mm = +1.5mm.

[0111] The second step is to determine the duty cycle level. The calculated spacing difference is input into a preset quantization level correspondence. Specifically, this quantization level correspondence can be a lookup table that maps the spacing difference range to discrete duty cycle levels. For example, if the spacing difference is known to be +1.5mm, and the table specifies that a difference of +1.0mm to +2.0mm corresponds to level 4, then duty cycle level 4 is generated.

[0112] The third step is to determine the signal conduction time percentage. This is done by looking up the corresponding signal conduction time percentage in a pre-defined correspondence based on the duty cycle level. For example, level 4 corresponds to a 65% conduction time percentage.

[0113] The fourth step is to generate a PWM signal. Based on the determined signal conduction time ratio of 65%, a preset fixed-cycle timing unit is filled. Within one pulse cycle, the first 65% of the time period is set to a high level, and the last 35% of the time period is set to a low level, thereby generating a PWM signal with a specific pulse width.

[0114] Finally, in step S1035, the pulse width modulation signal is transmitted to the stepper motor of the multi-channel adjustable-pitch optical attenuator via the drive circuit. The stepper motor then adjusts the spacing of the optical channels according to the signal pulse width. Finally, the near-infrared spectral data processed by the optical attenuator is acquired as standardized spectral data. For example, the pulse width modulation signal drives the stepper motor to precisely adjust the optical channel spacing to 3.5 mm. At this time, the intensity of the acquired spectral signal is controlled within the optimal detection range, forming standardized spectral data.

[0115] Specifically, on the raw material testing line of a biomass energy company, the system performs the following operations:

[0116] First, the most recent 30 seconds are divided into 30 consecutive 1-second time slices. Each time slice is traversed, and the particle size values ​​of all associated mapping data within it are extracted to form a set of particle size values ​​for each time slice. These values ​​are matched and accumulated with preset particle size intervals (e.g., 0-0.5mm, 0.5-1.0mm, 1.0-2.0mm) to calculate the distribution density for each interval. A confidence coefficient is calculated based on the stability of the distribution density, and a dynamic weighting factor is generated through the confidence weight mapping relationship. By comparing the weighted accumulated quantities, the 0.5-1.0mm interval is determined as the first particle size distribution interval.

[0117] Next, the identifier "DOM001" is assigned to this interval as the dominant interval identifier. Then, the preset interval spacing mapping table is queried using this identifier to obtain the target spacing value of 3.5mm.

[0118] Then, the value is encoded: the difference between it and the reference spacing (2.0 mm) is calculated (+1.5 mm); the duty cycle level is determined according to the quantization level correspondence (e.g., level 4); the corresponding signal conduction time percentage is found (e.g., 65%); and a pulse width modulation signal is generated accordingly.

[0119] Finally, the signal is loaded onto the actuator of the optical attenuator, driving it to adjust the channel spacing to 3.5 mm, thereby acquiring standardized near-infrared spectral data that eliminates particle size interference, completing the entire optical adaptive adjustment process based on real-time particle size feedback.

[0120] The above-mentioned S103 overall scheme effectively overcomes the interference of biomass raw material particle size variation on spectral detection by analyzing the raw material particle size distribution characteristics in real time and dynamically adjusting the optical system parameters, ensuring the comparability of spectral data of raw materials with different particle sizes, and providing a stable and reliable data foundation for subsequent quantitative analysis of components.

[0121] S104. Assign differentiated weights to the characteristic bands corresponding to the main components of the pre-set biomass raw materials to the standardized spectral data to generate enhanced characteristic spectral data.

[0122] Optionally, step S104 may specifically include the following steps:

[0123] S1041. Divide the standardized spectral data into multiple spectral data segments according to the wavelength of the standardized spectral data.

[0124] Among them, the spectral data segment refers to the standardized spectral data subset obtained by dividing by wavelength, including the spectral intensity value sequence within a specific wavelength range, which is used for differentiated processing based on wavelength intervals.

[0125] S1042. Match each of the spectral data segments with the characteristic bands corresponding to each of the main components of the pre-set biomass raw material, and identify the spectral data segments belonging to each of the characteristic bands from the matching results.

[0126] S1043. Multiply each of the spectral data segments and the weight values ​​corresponding to each of the preset weight value sets by the weight values ​​of each spectral data segment to obtain a weighted spectral data segment.

[0127] The preset weight value set refers to the pre-defined combination of weight coefficients, including the enhancement coefficients corresponding to each characteristic band and the reference coefficients of non-characteristic bands, which are used to implement spectral enhancement according to the importance of chemical components.

[0128] S1044. Multiply the spectral data segments that do not belong to each of the characteristic bands with the preset reference weight value to obtain the reference spectral data segments.

[0129] The preset benchmark weight value refers to the uniform weighting coefficient used for non-characteristic bands, including adjustment parameters that maintain or weaken the spectral signal, used to balance the spectral intensity distribution across the entire spectrum. The benchmark spectral data segment refers to the spectral data segment processed by the benchmark weight, including the adjusted spectral intensity values ​​of non-characteristic bands, used to maintain the integrity of the spectral data.

[0130] S1045. The weighted spectral data segments and the reference spectral data segments are recombined according to the wavelength order of the weighted spectral data segments and the wavelength order of the reference spectral data segments to generate enhanced feature spectral data.

[0131] The weighted spectral data segment refers to the spectral data segment processed by weighting coefficients, including magnified or reduced spectral intensity values, used to highlight the characteristic spectral information of specific chemical components. The enhanced characteristic spectral data refers to the recombined spectral data, including full-spectrum spectral information after weighted and baseline processing, used to highlight the characteristics of major components and suppress irrelevant signals.

[0132] In this embodiment of the application, firstly, through step S1041, the complete spectrum is divided into multiple continuous spectral data segments according to the wavelength range of the standardized spectral data and a preset wavelength interval. For example, the spectral range of 1100-2500nm is divided into 140 spectral data segments at 10nm intervals, and each data segment contains all absorbance data points within that wavelength range.

[0133] Secondly, in step S1042, the center wavelength of each spectral data segment is compared with the characteristic band range of each major component of the pre-set biomass raw material. Then, the spectral data segment falling within the characteristic band range is identified by the interval matching algorithm. For example, the spectral data segment is matched with the characteristic band of cellulose 1340-1360nm, and the spectral data segment with a center wavelength of 1350nm is identified as belonging to this characteristic band.

[0134] Subsequently, in step S1043, the weighting coefficients corresponding to each spectral data segment are obtained from the preset set of weighting values. Then, the spectral intensity value of each spectral data segment is multiplied by the corresponding weighting coefficient to obtain the weighted spectral data segment. For example, a weighting coefficient of 2.0 is applied to the spectral data segment belonging to the characteristic band of cellulose, thus doubling its spectral intensity value.

[0135] Next, a preset reference weight value is determined through step S1044. Then, the spectral data segments that do not belong to any characteristic band are multiplied by the reference weight value to obtain the reference spectral data segments that maintain the original intensity relationship. For example, a reference weight value of 0.8 is uniformly applied to the spectral data segments of non-characteristic bands to appropriately reduce their spectral intensity.

[0136] Finally, in step S1045, all weighted spectral data segments and reference spectral data segments are arranged in wavelength order, and then adjacent data segments are smoothly connected to generate complete enhanced feature spectral data. For example, spectral data segments processed with different weights are recombined in order of wavelength from low to high to form enhanced full-spectrum spectral data.

[0137] Specifically, in the quality control laboratory of a biomass chemical plant, the standardized near-infrared spectrum was first divided into multiple spectral data segments at 20nm intervals; then, spectral data segments belonging to the characteristic bands of cellulose, hemicellulose, and lignin were identified through band matching; then, weighted spectral data segments were obtained by applying weighting coefficients of 2.5, 2.0, and 1.8 to the characteristic band data segments respectively; at the same time, a reference weight of 0.7 was applied to the non-characteristic band data segments to obtain the reference spectral data segments; finally, all data segments were recombined in wavelength order to generate enhanced characteristic spectral data, thus completing the spectral feature enhancement processing.

[0138] The overall scheme of S104 described above effectively enhances the characteristic spectral signals of the main components of biomass by implementing differentiated weight allocation for different spectral bands, while maintaining the integrity of spectral data and significantly improving the correlation between spectral features and component content.

[0139] S105. Establish a multi-output regression model using the pre-set spectral data of biomass samples with known component contents.

[0140] In the above scheme, the multi-output regression model refers to a machine learning model that can simultaneously predict multiple target variables, including a shared hidden layer structure and multiple independent output layer nodes, which is used to synchronously output the predicted content values ​​of multiple components such as cellulose, hemicellulose and lignin based on the input spectral data.

[0141] In this embodiment of the application, firstly, the actual contents of cellulose, hemicellulose and lignin in biomass samples with known component contents are determined by laboratory standard methods, and these content data are established in a one-to-one correspondence with the near-infrared spectral data of the corresponding samples to form a training dataset containing spectral features and true component contents.

[0142] Secondly, a regression model architecture with multiple output nodes is constructed, where each output node corresponds to the content prediction of a component to be measured. At the same time, the input layer dimension of the model is designed to match the feature dimension of the spectral data, and the connection weights and bias parameters of the model are initialized.

[0143] Finally, an iterative optimization algorithm is used to input the training dataset into the model for multiple rounds of training. In each iteration, the error between the predicted value and the true value is calculated, and the internal parameters of the model are automatically adjusted according to the magnitude of the error until the model can stably and accurately predict the content of multiple components at the same time, and finally the trained multi-output regression model is obtained.

[0144] Specifically, in the analytical laboratory of a biomass energy company, 200 biomass samples with standard component content test reports and their corresponding near-infrared spectral data were first collected; then, a neural network model architecture with three output nodes was constructed, corresponding to the content prediction of cellulose, hemicellulose and lignin respectively; finally, the sample data was input into the model through batch training, and after multiple rounds of parameter optimization, a multi-output regression model that can accurately predict the content of the three components was obtained.

[0145] The overall solution of S105 described above, by establishing a multi-output regression model, achieves simultaneous and rapid prediction of the content of multiple components in biomass raw materials, avoiding the complexity and inconsistency of establishing multiple single-output models separately, and significantly improving the efficiency and accuracy of component detection.

[0146] S106. Input the enhanced characteristic spectral data into the multi-output regression model to generate the content data of multiple components in the pre-set biomass raw material.

[0147] Optionally, step S106 may specifically include the following steps:

[0148] S1061. Extract wavelength intensity values ​​from the enhanced feature spectral data.

[0149] Among them, wavelength intensity value refers to the signal intensity value of each wavelength point extracted from the enhanced characteristic spectral data, including optical parameters such as absorbance value or reflectance value, which is used to characterize the spectral response characteristics of biomass raw materials at different wavelengths.

[0150] S1062. Arrange and combine the wavelength intensity values ​​according to the wavelength order of the wavelength intensity values ​​to generate a spectral feature vector.

[0151] Among them, the spectral feature vector refers to the mathematical vector composed of intensity values ​​arranged in wavelength order, including an ordered intensity data sequence within the entire spectral range, which is used as standardized input data for the multi-output regression model.

[0152] S1063. Multiply the spectral feature vector and the parameter matrix in the multi-output regression model to generate an initial component content vector.

[0153] The initial component content vector refers to the preliminary prediction result obtained through linear transformation of the parameter matrix, including the initial estimated value of each component content, which is used to establish a preliminary correlation between spectral features and component content through linear relationship.

[0154] S1064. Add the initial component content vector and the bias vector in the multi-output regression model to generate an intermediate component content vector.

[0155] The intermediate component content vector refers to the predicted component content after bias correction, including the component content data adjusted to the baseline, which is used to improve the accuracy and stability of the prediction model.

[0156] S1065. Perform a nonlinear transformation on the intermediate component content vector to generate the final component content vector.

[0157] The final component content vector refers to the final result of the component content after nonlinear transformation, including the predicted value of component content that conforms to the actual distribution law, and is used to output the standardized prediction result after processing by the complete model.

[0158] S1066. Convert each element value in the final component content vector into the content data of each component in the pre-set biomass raw material corresponding to each element value.

[0159] In this embodiment of the application, firstly, through step S1061, the absorbance or reflectance values ​​corresponding to each wavelength point in the enhanced characteristic spectral data are read using the data parsing module to obtain a complete wavelength intensity value sequence. For example, the absorbance value of each nanometer wavelength point in the range of 1200nm to 2400nm is extracted from the enhanced characteristic spectral data to form an original data set containing 1201 intensity values.

[0160] Secondly, in step S1062, all wavelength intensity values ​​are sorted in ascending order of wavelength, and then the sorted intensity values ​​are combined sequentially into a numerical sequence with a fixed dimension to finally generate a spectral feature vector. For example, wavelength intensity values ​​from 1200nm to 2400nm are arranged at 1nm intervals and combined into a 1201-dimensional spectral feature vector for model input.

[0161] Subsequently, in step S1063, the spectral feature vector is multiplied by the parameter matrix trained in the multi-output regression model. A linear transformation maps the high-dimensional spectral features to the component content space, generating an initial component content vector containing preliminary predicted values ​​for each component. For example, multiplying the 1201-dimensional spectral feature vector by the parameter matrix yields a three-dimensional initial component content vector containing preliminary predicted values ​​for cellulose, hemicellulose, and lignin.

[0162] Next, in step S1064, the trained and optimized bias vector is obtained from the multi-output regression model. Then, the bias vector is added element-wise to the initial component content vector to perform benchmark correction on the prediction result, generating a more accurate intermediate component content vector. For example, the corresponding bias value is added to each element value of the initial component content vector to obtain the corrected intermediate component content vector.

[0163] Then, in step S1065, a nonlinear activation function is applied to each element value in the intermediate component content vector to transform the linear prediction result to a reasonable numerical range, and then the final component content vector that conforms to the actual content distribution characteristics is output. For example, the Sigmoid function is used to convert the predicted values ​​in the intermediate component content vector into percentage values ​​between 0 and 1 to generate the final component content vector.

[0164] Finally, in step S1066, each element value in the final component content vector is multiplied by the actual content scaling coefficient to convert the normalized value into a specific component content percentage. Then, the final content data of each component is output according to the preset component order. For example, the three element values ​​in the final component content vector are multiplied by 100 respectively to convert them into specific values ​​for cellulose content percentage, hemicellulose content percentage, and lignin content percentage.

[0165] Specifically, in the raw material quality inspection system of a biomass power plant, the intensity values ​​of all wavelength points are first extracted from the enhanced feature spectral data; then, these intensity values ​​are arranged and combined into a spectral feature vector according to the wavelength order; next, this vector is multiplied by the trained parameter matrix to obtain the initial component content vector; then, the initial vector is added to the bias vector to generate an intermediate component content vector; then, the final component content vector is obtained through a nonlinear function transformation; finally, the element values ​​in the vector are converted into specific component content percentages, completing the complete prediction process from spectral data to component content.

[0166] The overall scheme of S106 described above achieves a precise mathematical mapping from spectral features to component content through systematic vector operations and nonlinear transformation processing, ensuring the accuracy and reliability of the prediction of multiple component contents and providing an effective technical means for the rapid quantitative analysis of biomass raw material components.

[0167] The following is a complete example of steps 101-106: A large-scale biomass power plant uses near-infrared spectroscopy online detection to monitor the content of cellulose, hemicellulose, and lignin in sawdust raw materials during the transportation process in real time. First, in the raw material pretreatment workshop of a large-scale biomass power plant, a laser particle size analyzer installed on the conveyor belt scans the flowing sawdust raw materials in real time to obtain statistical data on the distribution of multiple particle size ranges. At the same time, a near-infrared spectral probe is used to scan the same raw material flow across the entire spectrum, collecting spectral intensity information covering the characteristic wavelength range, forming the original particle size distribution data and near-infrared spectral dataset.

[0168] Secondly, the data acquisition module of the central control system adds precise timestamps to all input particle size distribution data and near-infrared spectral data. Through a sliding time window algorithm, data points at the same time are paired and associated to generate an association mapping data table that reflects the correspondence between particle size characteristics and spectral response, providing a synchronous and complete data foundation for subsequent analysis.

[0169] Subsequently, the system performs real-time analysis of the associated mapping data, determines the dominant particle size range by statistically analyzing the frequency of particle size values ​​within the time slice, and obtains the target spacing value by querying the preset mapping table based on the interval identifier. Then, the value is encoded into a pulse width modulation signal to drive the stepper motor of the optical attenuator, automatically adjusting the optical channel spacing, thereby outputting near-infrared spectral data with normalized light intensity.

[0170] Then, the processing unit divides the standardized spectral data into multiple spectral data segments according to wavelength intervals, identifies specific data segments belonging to the characteristic bands of cellulose, hemicellulose, and lignin through a matching algorithm, and applies differential weight coefficients to them to enhance their intensity. At the same time, a baseline weight value is applied to the non-characteristic band data segments. Finally, all data segments are recombined to generate enhanced characteristic spectral data that highlights the component characteristics.

[0171] Next, the laboratory used an accumulated biomass sample library with known component contents, including wood and crop residue samples from different sources, to build a multi-output regression model through batch training. This model can simultaneously learn the complex mapping relationship between spectral features and the contents of multiple components, and optimize internal parameters to achieve accurate prediction.

[0172] Finally, in the actual testing process, the real-time generated enhanced feature spectral data is input into the trained multi-output regression model. After processing steps such as feature vector extraction, matrix operation, bias correction and nonlinear transformation, the percentage results of cellulose content, hemicellulose content and lignin content are directly output, completing the rapid online quantitative analysis of biomass raw material components.

[0173] Figure 3 This is a schematic diagram of a specific embodiment of an online near-infrared spectroscopy detection system for biomass raw material components provided in this application. (Refer to...) Figure 3 The system may include:

[0174] The acquisition module 31 is used to acquire particle size distribution data and near-infrared spectral data of the pre-set biomass raw materials;

[0175] Processing module 32 is used to perform time stamp alignment processing on the particle size distribution data and the near-infrared spectral data to obtain the correlation mapping data between the particle size distribution data and the near-infrared spectral data;

[0176] The adjustment module 33 uses the associated mapping data to adjust the spacing of each channel in the preset multi-channel adjustable optical attenuator to obtain standardized spectral data.

[0177] The allocation module 34 allocates differentiated weights to the characteristic bands corresponding to the main components of the pre-set biomass raw materials to the standardized spectral data in order to generate enhanced characteristic spectral data.

[0178] Module 35 is established to build a multi-output regression model using pre-set spectral data of biomass samples with known component contents;

[0179] The generation module 36 inputs the enhanced characteristic spectral data into the multi-output regression model to generate the content data of multiple components in the preset biomass raw material.

[0180] This application provides an online near-infrared spectroscopy detection system for biomass raw material components to implement the aforementioned online near-infrared spectroscopy detection method for biomass raw material components. Therefore, the specific implementation of the online near-infrared spectroscopy detection system for biomass raw material components can be found in the embodiment section of the online near-infrared spectroscopy detection method for biomass raw material components described above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0181] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described method for online detection of near-infrared spectrometry of biomass raw material components.

[0182] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for online near-infrared spectroscopy detection of biomass raw material components.

[0183] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0184] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in the embodiments of the online near-infrared spectroscopy detection method for any of the above-described biomass raw material components.

[0185] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0186] The above provides a detailed description of the near-infrared spectroscopy online detection method and system for biomass raw material components provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A near-infrared spectroscopy online detection method for components of biomass raw materials, characterized in that, include: Obtain particle size distribution data and near-infrared spectral data of the pre-placed biomass raw materials; The particle size distribution data and the near-infrared spectral data are time-stamp aligned to obtain the correlation mapping data between the particle size distribution data and the near-infrared spectral data; The spacing between each channel in the preset multi-channel adjustable optical attenuator is adjusted using the aforementioned correlation mapping data to obtain standardized spectral data; Differential weights corresponding to the characteristic bands of each major component of the pre-set biomass raw material are assigned to the standardized spectral data to generate enhanced characteristic spectral data; A multi-output regression model was established using spectral data of biomass samples with known component contents. The enhanced characteristic spectral data is input into the multi-output regression model to generate content data of multiple components in the pre-set biomass raw material.

2. The method according to claim 1, characterized in that, The process of adjusting the spacing of each channel in a preset multi-channel adjustable-distance optical attenuator using the associated mapping data to obtain standardized spectral data includes: The correlation mapping data is analyzed to determine the particle size distribution interval with the highest frequency of particle size values ​​in the correlation mapping data as the first particle size distribution interval; The first particle size distribution range is identified to obtain the dominant range identifier; The target spacing value is obtained by accessing the preset interval spacing mapping table based on the dominant interval identifier. The target spacing value is encoded to generate a pulse width modulation signal; The pulse width modulation signal is loaded into the actuator of a preset multi-channel adjustable optical attenuator to generate near-infrared spectral data as standardized spectral data.

3. The method according to claim 2, characterized in that, The step of analyzing the correlation mapping data and determining the particle size distribution interval with the highest frequency of particle size values ​​in the correlation mapping data as the first particle size distribution interval includes: The preset statistical time window is divided into multiple consecutive time slices to determine the time slice sequence of the current statistical period; Traverse each time slice in the time slice sequence and extract the granularity value of each associated mapping data in each time slice to generate a set of granularity values ​​for each time slice. The set of particle size values ​​and the preset particle size intervals to which each set of particle size values ​​belongs are matched, and a cumulative quantity corresponding to each preset particle size interval is generated based on the matching result. The ratio between the cumulative quantity of each particle size and the total number of particle size values ​​in each set of particle size values ​​is calculated, and the ratio calculation result is used as the distribution density of each preset particle size interval. Calculate the confidence coefficient of the distribution characteristics of each particle size value in the current statistical period based on the distribution density; Dynamic weighting is applied to the cumulative quantities based on the confidence coefficients to generate the particle size distribution interval with the highest frequency of particle size values ​​as the first particle size distribution interval.

4. The method according to claim 3, characterized in that, The step of dynamically weighting the cumulative quantities according to the confidence coefficients to generate the particle size distribution interval with the highest frequency of particle size values ​​as the first particle size distribution interval includes: Each confidence coefficient is input into a preset confidence weight mapping relationship to generate a dynamic weight factor corresponding to each preset particle size interval to which each set of particle size values ​​belongs. Each of the cumulative quantities is multiplied by the corresponding dynamic weighting factor to obtain the weighted cumulative quantity for each of the preset particle size intervals; The numerical values ​​of each weighted cumulative quantity are compared, and the preset particle size interval corresponding to the weighted cumulative quantity with the largest value is taken as the particle size distribution interval with the highest frequency of particle size value occurrence, and is taken as the first particle size distribution interval.

5. The method according to claim 2, characterized in that, Encoding the target spacing value to generate a pulse width modulation signal includes: The target spacing value and the preset reference spacing value are used to perform a difference calculation, and the result of the difference calculation is used as the spacing difference. The spacing difference is input into a preset quantization level correspondence to generate a duty cycle level corresponding to the spacing difference; The signal conduction time percentage corresponding to the duty cycle level is determined from the preset correspondence between duty cycle level and signal conduction time percentage. The preset pulse period timing unit is filled according to the signal conduction time ratio to generate a pulse width modulation signal.

6. The method according to claim 1, characterized in that, The step of assigning differentiated weights to the characteristic bands corresponding to each major component of the pre-set biomass raw material to the standardized spectral data to generate enhanced characteristic spectral data includes: The standardized spectral data is divided into multiple spectral data segments based on the wavelength of the standardized spectral data. Each of the spectral data segments is matched with the characteristic bands corresponding to each major component of the pre-set biomass raw material, and the spectral data segments belonging to each of the characteristic bands are identified from the matching results. The weighted spectral data segments are obtained by multiplying each spectral data segment with the weight value corresponding to each spectral data segment from the preset set of weight values. The spectral data segments that do not belong to each of the aforementioned characteristic bands are multiplied by a preset reference weight value to obtain a reference spectral data segment. The weighted spectral data segments and the reference spectral data segments are recombined according to the wavelength order of the weighted spectral data segments and the wavelength order of the reference spectral data segments to generate enhanced feature spectral data.

7. The method according to claim 1, characterized in that, The step of inputting the enhanced characteristic spectral data into the multi-output regression model to generate content data of multiple components in the pre-set biomass raw material includes: Extract wavelength intensity values ​​from the enhanced spectral data; The wavelength intensity values ​​are arranged and combined according to the wavelength order to generate a spectral feature vector; The initial component content vector is generated by multiplying the spectral feature vector and the parameter matrix in the multi-output regression model. The initial component content vector and the bias vector in the multi-output regression model are added together to generate the intermediate component content vector. The intermediate component content vector is nonlinearly transformed to generate the final component content vector. Each element value in the final component content vector is converted into the content data of each component in the pre-set biomass raw material corresponding to each element value.

8. A near-infrared spectroscopy online detection system for biomass raw material components, characterized in that, include: The acquisition module is used to acquire particle size distribution data and near-infrared spectral data of the pre-set biomass raw materials; The processing module is used to perform time stamp alignment processing on the particle size distribution data and the near-infrared spectral data to obtain the correlation mapping data between the particle size distribution data and the near-infrared spectral data; The adjustment module uses the associated mapping data to adjust the spacing of each channel in the preset multi-channel adjustable optical attenuator to obtain standardized spectral data. The allocation module assigns differentiated weights to the characteristic bands corresponding to the main components of the pre-set biomass raw materials to the standardized spectral data in order to generate enhanced characteristic spectral data. A module is established to build a multi-output regression model using spectral data of biomass samples with known component contents; The generation module inputs the enhanced characteristic spectral data into the multi-output regression model to generate the content data of multiple components in the preset biomass raw materials.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the near-infrared spectroscopy online detection method for biomass raw material components as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables an online near-infrared spectral detection method for biomass raw material components as described in any one of claims 1 to 7.

Images