A rapid catalytic evaluation method for lignocellulosic biomass performance

Abstract

The application discloses a kind of quick catalytic evaluation methods of lignocellulosic biomass performance, including making lignocellulosic biomass into powder and detecting inherent property parameters;Quick catalytic evaluation system is constructed, and reaction time is 1 / 10-1 / 5 of industrialization;Detection key performance indicators;With inherent property as input, performance index as output, adopt multilayer perceptron neural network to construct prediction model;Input raw material parameters to be screened to obtain predicted value, and determine adaptability according to threshold value.The application realizes the quantitative detection and prediction of lignocellulosic biomass performance, selects the most suitable lignocellulosic biomass raw material for target pretreatment process, improves the detection and screening efficiency, reduces production cost, and promotes the precision and efficient development of lignocellulosic biomass raw material analysis and detection industry.

Description

Technical Field

[0001] This invention belongs to the field of lignocellulose biomass detection and analysis technology, and particularly relates to a rapid catalytic evaluation method for the performance of lignocellulose biomass. Background Technology

[0002] Lignocellulosic biomass is the most abundant renewable biological resource on Earth, characterized by its diversity, numerous components, and complex structure. It is extremely common in daily life, including timber such as poplar and willow, grasses such as reeds and pennisetum, agricultural and forestry wastes such as corn stalks, wheat stalks, rice husks, bark, sawdust, and leaves, industrial wastes such as distiller's grains, cardboard boxes, and newspapers, and animal manure. Compared to traditional fossil fuels like coal, using lignocellulosic biomass as a raw material for producing fuels, chemicals, and materials offers numerous advantages, such as renewability, bioconversion capability, low sulfur and nitrogen content, and wide distribution. The detection and screening of lignocellulosic biomass raw materials is a core pre-processing step in industries such as lignocellulosic resource utilization and engineered wood product manufacturing. Its purpose is to accurately determine the suitability of raw materials by analyzing their composition, structure, and reaction characteristics, laying the foundation for subsequent pretreatment, catalytic conversion, enzymatic hydrolysis, and saccharification processes. Raw material testing and screening, as the first step in the utilization of lignocellulose biomass, directly determines pretreatment efficiency, product quality, and production costs, and is a key prerequisite for achieving precise and green processes.

[0003] Currently, existing screening methods for lignocellulosic biomass raw materials mainly remain at the property description level. That is, they preliminarily classify the raw materials by detecting their basic physicochemical properties, and then combine this with experience to determine whether they are suitable for a specific pretreatment process. Traditional screening methods can only remove impurities or simply distinguish between differences in appearance and density of raw materials, and cannot achieve accurate detection and prediction of the reactivity of raw materials. In the field of biorefining, most methods only roughly determine applicability by detecting lignin and cellulose content, and it is difficult to correlate the properties of raw materials with core detection indicators such as catalytic conversion efficiency and sugar yield after pretreatment.

[0004] The above methods have significant drawbacks: First, they have a single detection dimension, focusing only on the inherent properties of the raw materials themselves and ignoring the dynamic reaction characteristics of the raw materials in the catalytic environment, thus failing to reflect the actual detection performance of the raw materials after pretreatment. Second, they have poor predictability, only providing qualitative descriptions and failing to quantitatively detect and predict key parameters such as reaction efficiency and product yield of the raw materials under specific processes. Third, they have long detection cycles and high costs, with some fine detection steps being cumbersome and unable to provide rapid feedback, making them difficult to adapt to the needs of efficient industrial screening. Fourth, they are difficult to adapt to new pretreatment technologies. With the development of new technologies such as precision catalysis and green solvents, the requirements for the accuracy of raw material detection and screening are higher, and traditional methods can no longer meet the needs of industrial upgrading.

[0005] Therefore, there is an urgent need for an intelligent, precise, and efficient method for the detection and screening of lignocellulose biomass raw materials, to establish the correlation between the inherent properties of the raw materials and the catalytic performance of pretreatment, and to achieve the precision, speed, and quantification of the analysis, detection, and screening of lignocellulose biomass raw materials. Summary of the Invention

[0006] To address the technical problems of existing lignocellulosic biomass evaluation methods, which can only provide "property descriptions," cannot achieve performance testing and quantitative prediction, have low screening accuracy, and poor adaptability, this invention provides a rapid catalytic evaluation method for lignocellulosic biomass performance. By constructing a rapid catalytic evaluation system, it correlates the inherent properties of raw materials with the catalytic detection performance of pretreatment, enabling quantitative detection and prediction of raw material pretreatment performance. This allows for the screening of lignocellulosic biomass raw materials most suitable for the target pretreatment process, improving detection and screening efficiency, reducing production costs, and promoting the precise and efficient development of the lignocellulosic biomass raw material analysis and testing industry.

[0007] Based on this, the present invention provides a rapid catalytic evaluation method for the properties of lignocellulose biomass, the specific technical solution of which is as follows:

[0008] A rapid catalytic evaluation method for the properties of lignocellulose biomass includes the following steps:

[0009] S1 Select lignocellulose biomass raw material, remove impurities, crush and sieve to obtain lignocellulose biomass powder sample with uniform particle size; dry the lignocellulose biomass powder sample to remove moisture and set aside for later use.

[0010] S2 Detects the inherent property parameters of the lignocellulose biomass powder sample, including: cellulose content, hemicellulose content, lignin content, cellulose crystallinity, specific surface area, and porosity;

[0011] S3. Based on the target pretreatment process, determine the corresponding rapid catalytic evaluation conditions to construct a rapid catalytic detection and evaluation system; the rapid catalytic evaluation conditions include: catalytic reaction temperature, catalytic reaction time, catalyst type and dosage, reaction medium and concentration, and solid-liquid ratio; the catalytic reaction time is 1 / 10 to 1 / 5 of the actual industrial pretreatment reaction time;

[0012] S4 Take the lignocellulose biomass powder sample and place it into the rapid catalytic detection and evaluation system of S3 for catalytic reaction; after the reaction is completed, separate the reaction products and detect the key performance indicators of the reaction products; the key performance indicators include: lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, target product yield and inhibitor generation amount;

[0013] S5 uses the inherent property parameters of the lignocellulosic biomass powder sample detected in S2 as input variables and the key performance indicators detected in S4 as output variables to construct a raw material performance prediction model using a multilayer perceptron artificial neural network.

[0014] S6 The lignocellulosic biomass raw material to be screened is processed and tested according to the methods described in S1 to S2 to obtain its inherent property parameters; the inherent property parameters are input into the raw material performance prediction model constructed in S5, and the model outputs the performance index prediction value of the lignocellulosic biomass raw material to be screened under the target pretreatment process; a screening threshold is set, and if the performance index prediction value meets the screening threshold, the raw material is determined to be a suitable raw material; if it does not meet the threshold, it is determined to be an unsuitable raw material.

[0015] Preferably, in S3, the target pretreatment process includes one of alkaline hydrogen peroxide catalytic pretreatment, solid acid catalytic pretreatment, ionic liquid synergistic catalytic pretreatment, and mechanical-enzyme synergistic catalytic pretreatment.

[0016] Preferably, in S1, the pulverization is carried out using a high-speed universal pulverizer, the sieving is carried out using a standard inspection sieve, and the particle size of the lignocellulose biomass powder sample is 40-100 mesh; the drying is carried out using a vacuum drying oven, the drying temperature is 60-80℃, the drying time is 2-4h, and the moisture content after drying is ≤8wt%.

[0017] Preferably, in S2, the cellulose content and hemicellulose content are determined by the Paraná separation method, the lignin content is determined by the Klason method, the cellulose crystallinity is determined by X-ray diffraction, and the specific surface area and porosity are determined by nitrogen adsorption-desorption method.

[0018] Preferably, in S4, the catalytic reaction is carried out using a constant temperature water bath shaker or a high-pressure reactor, and the stirring speed is controlled at 100-200 r / min during the reaction; the reaction products are separated by centrifugation, with a centrifugation speed of 3000-5000 r / min and a centrifugation time of 5-10 min.

[0019] Preferably, in S5, the multilayer perceptron artificial neural network is a feedforward fully connected network, including an input layer, a hidden layer, and an output layer; the input layer has neurons corresponding to the inherent property parameters of the lignocellulosic biomass powder sample; the output layer has neurons corresponding to the key performance indicators; and the hidden layer is set to 1-3 layers according to the target pretreatment process.

[0020] Preferably, the input layer uses min-max normalization to normalize the inherent property parameters of the lignocellulosic biomass powder samples in the training set, mapping them to the [0,1] interval; the activation function of the hidden layer uses a modified linear unit function or a modified linear unit function with leakage; the output layer uses a linear activation function to directly output the quantitative prediction results.

[0021] Preferably, the weights are initialized using the He normal initialization method, and the initial value of the bias term is set to 0.01; an adaptive moment estimation optimizer is used for training, with mean squared error as the loss function; a combined regularization strategy of L2 regularization and Dropout layer is used to prevent overfitting; the samples are randomly divided into training set and validation set in a 7:3 ratio, and the early stopping method is used to control the number of training rounds.

[0022] Preferably, the screening threshold in S6 is set according to the performance requirements of the target pretreatment process, including the minimum value of lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, target product yield, and the maximum value of inhibitor generation.

[0023] Preferably, the catalyst includes one or more of solid acid catalysts, metal complex catalysts, and enzyme catalysts; the reaction medium includes one or more of deionized water, dilute acid solution, dilute alkali solution, or ionic liquid, with a solid-liquid ratio of 1:10 g / mL to 1:20 g / mL.

[0024] The technical solution of the present invention has the following advantages:

[0025] 1. Achieve a leapfrog improvement in screening methods, breaking through the limitations of traditional "property descriptions," and establishing a precise correlation between the inherent properties of lignocellulosic biomass raw materials and the catalytic performance of pretreatment. This truly realizes "performance prediction," solving the core pain point that existing screening methods cannot predict the actual performance of raw materials after pretreatment. It significantly improves the accuracy and adaptability of raw material screening, avoiding the problems of low efficiency and excessive energy consumption caused by mismatch between raw materials and pretreatment processes.

[0026] 2. Construct a rapid catalytic evaluation system to shorten the catalytic reaction time to 1 / 10-1 / 5 of the actual industrial process. This eliminates the need for complex and lengthy detailed detection of raw material properties, significantly shortens the screening cycle, reduces screening costs, and meets the high-efficiency screening requirements of large-scale industrial production, thereby improving production efficiency.

[0027] 3. By using machine learning algorithms to build a performance prediction model, the prediction accuracy is greatly improved, enabling quantitative prediction of raw material pretreatment performance. The screening criteria are more objective and scientific, avoiding subjective errors caused by reliance on experience judgment in traditional screening methods, and improving the stability and reliability of screening results.

[0028] 4. It has wide adaptability and can flexibly adjust the rapid catalytic evaluation conditions and screening thresholds according to different target pretreatment processes (such as alkaline catalysis, solid acid catalysis, ionic liquid synergistic catalysis, etc.). It is suitable for raw material screening needs in multiple fields such as biorefining, artificial board production, and deep processing of lignocellulosic biomass. At the same time, it can be adapted to new pretreatment technologies to help upgrade industrial processes.

[0029] 5. The operation process is simple and highly repeatable. The steps of raw material pretreatment, property detection, and catalytic evaluation all adopt conventional equipment and standard methods, without the need for special high-end equipment. It is easy to promote and apply industrially and has strong practicality and industrialization value. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of a rapid catalytic evaluation method for the properties of lignocellulose biomass according to the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0033] Lignocellulose biomass is the most abundant renewable biological resource on Earth. It is the collective term for the cell walls of terrestrial plants, a natural composite polymer material composed primarily of cellulose, hemicellulose, and lignin, supplemented with ash, pectin, small amounts of protein, salts, and minerals. It is the core material supporting the structure of plants. Lignocellulose biomass has a wide range of sources, including forest trees (poplar, willow, etc.), grasses (reed, foxtail grass, etc.), agricultural and forestry waste (corn stalks, wheat straw, rice husks, bark, sawdust, leaves, etc.), and industrial waste (distillers' grains, cardboard boxes, newspapers, etc.). Furthermore, cellulose is a chain-like polymer whose structural unit is D-glucose, linked by β-1,4 glycosidic bonds. It is the most abundant organic component of plant cell walls and the largest component in major lignocellulose biomass. Cellulose structure contains crystalline and amorphous regions. In the crystalline region, the cellulose chains formed by the polymerization of glucose are arranged in a regular and orderly manner, with numerous hydrogen bonds between the bundled fibers. The amorphous region has a looser and more disordered structure. Therefore, the crystallinity of cellulose is also an important evaluation indicator. Hemicellulose has a much more complex composition, consisting mainly of low-molecular-weight polysaccharides polymerized from xylose, arabinose, glucose, galactose, mannose, fucose, glucuronic acid, and galacturonic acid. It is primarily classified into three categories: xylooligosaccharides, polyglucomannans, and polygalactoglucomannans. Lignin is a three-dimensional amorphous polymer composed of methoxyphenylpropane structures.

[0034] Please see first. Figure 1 This invention discloses a rapid catalytic evaluation method for the properties of lignocellulose biomass, specifically including the following steps:

[0035] S1 Select lignocellulose biomass raw material, remove impurities, crush and sieve to obtain lignocellulose biomass powder sample with uniform particle size; dry the lignocellulose biomass powder sample to remove moisture and set aside for later use.

[0036] S2 Detects the inherent property parameters of the lignocellulose biomass powder sample, including: cellulose content, hemicellulose content, lignin content, cellulose crystallinity, specific surface area, and porosity;

[0037] S3. Based on the target pretreatment process, determine the corresponding rapid catalytic evaluation conditions to construct a rapid catalytic detection and evaluation system; the rapid catalytic evaluation conditions include: catalytic reaction temperature, catalytic reaction time, catalyst type and dosage, reaction medium and concentration, and solid-liquid ratio; the catalytic reaction time is 1 / 10 to 1 / 5 of the actual industrial pretreatment reaction time;

[0038] S4 Take the lignocellulose biomass powder sample and place it into the rapid catalytic detection and evaluation system of S3 for catalytic reaction; after the reaction is completed, separate the reaction products and detect the key performance indicators of the reaction products; the key performance indicators include: lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, target product yield and inhibitor generation amount;

[0039] S5 uses the inherent property parameters of the lignocellulosic biomass powder sample detected in S2 as input variables and the key performance indicators detected in S4 as output variables to construct a raw material performance prediction model using a multilayer perceptron artificial neural network.

[0040] S6 The lignocellulosic biomass raw material to be screened is processed and tested according to the methods described in S1 to S2 to obtain its inherent property parameters; the inherent property parameters are input into the raw material performance prediction model constructed in S5, and the model outputs the performance index prediction value of the lignocellulosic biomass raw material to be screened under the target pretreatment process; a screening threshold is set, and if the performance index prediction value meets the screening threshold, the raw material is determined to be a suitable raw material; if it does not meet the threshold, it is determined to be an unsuitable raw material.

[0041] Specifically:

[0042] S1 Select lignocellulose biomass raw materials, remove impurities, and then crush and sieve them to obtain lignocellulose biomass powder samples with uniform particle size; dry the lignocellulose biomass powder samples to remove moisture, and set aside for later use.

[0043] This step is the raw material pretreatment step. Specifically, select the lignocellulosic biomass raw material to be screened, such as forest trees and timber, remove impurities (such as bark, metal, plastic, stones, etc.), and then crush and sieve it to obtain a lignocellulosic biomass powder sample with uniform particle size. The lignocellulosic biomass powder sample is then dried to remove moisture and set aside for later use. The crushing is done using a high-speed universal crusher, the sieving uses a standard test sieve, and the drying is done in a vacuum drying oven at a temperature of 60-80℃ for 2-4 hours to ensure uniform drying and avoid localized moisture residue.

[0044] The pulverization was carried out using a high-speed universal pulverizer, the sieving was carried out using standard inspection sieves, and the drying was carried out using a vacuum drying oven at a temperature of 60-80℃ for 2-4 hours to ensure uniform drying of the sample and avoid local moisture residue.

[0045] S2 detects the inherent property parameters of the lignocellulose biomass powder sample, including: cellulose content, hemicellulose content, lignin content, cellulose crystallinity, specific surface area, and porosity;

[0046] This step involves detecting the inherent properties of the lignocellulose biomass powder sample. The six parameters mentioned above (cellulose content, hemicellulose content, lignin content, cellulose crystallinity, specific surface area, and porosity) are the core fundamental parameters affecting the catalytic performance of lignocellulose biomass feedstock pretreatment: the cellulose, hemicellulose, and lignin contents determine the convertibility potential of the feedstock's chemical composition; cellulose crystallinity reflects the resistance to degradation of the feedstock's supramolecular structure; and specific surface area and porosity characterize the accessibility of the feedstock's microstructure to the catalyst. By detecting these six parameters, the pretreatment response characteristics of the feedstock can be comprehensively reflected from three dimensions: chemical composition, supramolecular structure, and physical morphology, providing multidimensional input features for subsequent performance prediction models.

[0047] The cellulose and hemicellulose contents were determined using the Pantheon separation method, the lignin content was determined using the Klason method, the cellulose crystallinity was determined using X-ray diffraction, and the specific surface area and porosity were determined using the nitrogen adsorption-desorption method. All detection methods comply with relevant national standards or industry-standard methods to ensure the comparability of the test data.

[0048] S3. Based on the target pretreatment process, determine the corresponding rapid catalytic evaluation conditions to construct a rapid catalytic detection and evaluation system; the rapid catalytic evaluation conditions include: catalytic reaction temperature, catalytic reaction time, catalyst type and dosage, reaction medium and concentration, and solid-liquid ratio; the catalytic reaction time is 1 / 10 to 1 / 5 of the actual industrial pretreatment reaction time;

[0049] This step involves constructing a rapid catalytic detection and evaluation system. The target lignocellulosic biomass pretreatment process refers to a pretreatment technology that requires catalytic conversion or decomposition of lignocellulosic biomass raw materials, specifically including but not limited to: alkaline hydrogen peroxide catalytic pretreatment, solid acid catalytic pretreatment, ionic liquid synergistic catalytic pretreatment, and mechanical-enzyme synergistic catalytic pretreatment. The common characteristic of these processes is that they all rely on a catalyst (chemical catalyst, enzyme catalyst, or synergistic catalytic system) acting on the lignocellulosic biomass raw materials to disrupt the lignin-carbohydrate complex (LCC) structure and improve the accessibility of cellulose and hemicellulose. The core differences between different processes lie in the type of catalyst, reaction medium, and reaction conditions, but all can be adapted by adjusting parameters such as catalytic reaction temperature, reaction time, catalyst type and dosage, reaction medium and concentration, and solid-liquid ratio in step S3 of this invention. Therefore, the detection and screening method of this invention has broad process adaptability and is not limited to the four processes listed above. Any lignocellulosic biomass pretreatment process that relies on catalysis can be applied using the screening method of this invention by adjusting the rapid catalytic evaluation conditions.

[0050] In addition, the catalyst includes one or more of the following: solid acid catalysts (such as sulfonated biochar, zeolite), metal complex catalysts (such as copper(II)-2,2'-bipyridine complex), and enzyme catalysts, which are adapted to different types of pretreatment catalytic processes; the reaction medium is selected according to the type of catalyst and can be one of deionized water, dilute acid solution, dilute alkali solution, or ionic liquid to ensure the smooth progress of the catalytic reaction.

[0051] Finally, the design principles of the rapid catalytic detection and evaluation system are as follows:

[0052] (1) Catalytic reaction temperature: Set according to the optimal reaction temperature range of the target pretreatment process to ensure that the catalytic reaction is carried out under effective thermodynamic conditions. The temperature selection should be consistent with or slightly higher than the actual industrial process (not exceeding 10°C) to accelerate the reaction process while ensuring the consistency of the reaction mechanism.

[0053] (2) Catalytic reaction time: This is one of the core parameters of this invention, set to 1 / 10 to 1 / 5 of the actual industrial pretreatment reaction time. This time-reduction design is based on the following considerations: Under the premise of ensuring that the catalytic reaction mechanism remains unchanged, the reaction kinetics process is accelerated by increasing the catalyst concentration, optimizing the solid-liquid ratio, or slightly increasing the reaction temperature, so that the rapid evaluation results can accurately reflect the reaction trend under actual process conditions. At the same time, this time reduction ratio has been verified by a large number of experiments, which can ensure that the key performance indicators (such as lignin removal rate, sugar yield, etc.) maintain a high correlation with the actual process results (correlation coefficient R). 2 ≥0.95).

[0054] It is worth noting that in this invention, the catalytic reaction time is set to 1 / 10 to 1 / 5 of the actual industrial pretreatment reaction time. This range has been determined through extensive experimental verification. When the time reduction ratio is less than 1 / 10 (i.e., the reaction time is shorter), although the screening efficiency is further improved, the reaction process of the rapid catalytic reaction differs too much from that of the industrial process. The correlation coefficients between key performance indicators such as lignin removal rate and sugar yield and the actual process results drop below 0.85, indicating insufficient prediction accuracy. When the time reduction ratio is greater than 1 / 5 (i.e., the reaction time is longer), although the prediction accuracy is improved, the screening cycle is prolonged, failing to demonstrate the advantages of "rapid evaluation" and reducing the value for industrial application. Within the range of 1 / 10 to 1 / 5, the correlation coefficients between key performance indicators and the actual process results can reach above 0.95, while the screening cycle is significantly shortened, achieving the optimal balance between evaluation efficiency and prediction accuracy.

[0055] Therefore, the time reduction ratio of 1 / 10 to 1 / 5 is a key technical parameter determined through experimental optimization in this invention, and is not a simple ratio selection.

[0056] (3) Catalyst type and dosage: The type of catalyst shall be consistent with the actual industrial pretreatment process, including but not limited to one or more of the following: solid acid catalysts (such as sulfonated biochar, zeolite), metal complex catalysts (such as copper(II)-2,2'-bipyridine complex), and enzyme catalysts. The dosage of catalyst may be appropriately increased according to the time reduction requirements, but shall not exceed twice the actual dosage, so as to avoid changing the reaction pathway.

[0057] (4) Reaction medium and concentration: Select a suitable reaction medium according to the type of catalyst. One of the following can be used: deionized water, dilute acid solution (such as 0.5-2wt% sulfuric acid), dilute alkali solution (such as 1-5wt% sodium hydroxide), or ionic liquid. The concentration of the medium should be consistent with the actual process to ensure the consistency of the reaction environment.

[0058] (5) Solid-liquid ratio: set to 1:10 to 1:20 (g / mL). This range ensures that the lignocellulose biomass powder sample is fully dispersed in the reaction medium, avoiding the influence of mass transfer limitation on the reaction rate. The solid-liquid ratio can be adjusted according to the actual process, but the homogeneity and repeatability of the reaction system must be ensured.

[0059] By synergistically setting the above conditions, the constructed rapid catalytic detection and evaluation system can highly simulate the core catalytic environment of the target pretreatment process while significantly shortening the reaction time, enabling the rapid catalytic reaction test results to effectively predict the performance of actual industrial processes.

[0060] S4 Take the lignocellulose biomass powder sample and place it into the rapid catalytic detection and evaluation system of S3 for catalytic reaction; after the reaction is completed, separate the reaction products and detect the key performance indicators of the reaction products; the key performance indicators include: lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, target product yield and inhibitor generation amount;

[0061] This step involves rapid catalytic reaction testing to obtain actual reaction performance data of the raw materials in the catalytic environment, which serves as the output variable of the model. Key performance indicators include: lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, target product yield, and inhibitor formation. These five indicators constitute a complete pretreatment effect evaluation system: lignin removal rate evaluates the barrier breaking effect; cellulose and hemicellulose conversion rates evaluate the component utilization rate; target product yield (such as glucose, xylose, etc.) evaluates the economic output value; and inhibitor formation (such as phenolic compounds, furfural) evaluates the by-product risk. These five indicators complement each other, comprehensively reflecting the reaction performance of the raw materials in the pretreatment catalytic environment from four dimensions: removal efficiency, conversion degree, product yield, and by-product risk. These indicators directly reflect the reaction performance of the raw materials in the pretreatment catalytic environment and are the core basis for performance prediction. The measurement process employs precise detection methods such as high-performance liquid chromatography and ultraviolet spectrophotometry to ensure data reliability. Of course, other indicators can also be selected as key performance indicators according to actual needs.

[0062] Furthermore, in step S4, the catalytic reaction is carried out using a constant-temperature water bath shaker or a high-pressure reactor. During the reaction, the stirring speed is controlled at 100-200 r / min to ensure uniform suspension of the lignocellulose biomass powder, preventing sedimentation and foaming, ensuring reaction homogeneity, and facilitating a complete catalytic reaction. The reaction products are separated by centrifugation at 3000-5000 r / min for 5-10 min, achieving rapid and thorough solid-liquid separation and meeting the sample clarity requirements for subsequent testing. All the above parameters have been experimentally optimized to achieve rapid testing while ensuring data quality.

[0063] S5 uses the inherent property parameters of the lignocellulosic biomass powder sample detected in S2 as input variables and the key performance indicators detected in S4 as output variables to construct a raw material performance prediction model using a multilayer perceptron artificial neural network.

[0064] This step involves constructing the performance prediction model. The multilayer perceptron artificial neural network is a feedforward fully connected network, comprising an input layer, hidden layers, and an output layer. The input layer has six neurons, corresponding to cellulose content, hemicellulose content, lignin content, cellulose crystallinity, specific surface area, and porosity, respectively. The output layer has five neurons, corresponding to lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, target product yield, and inhibitor generation, respectively. Of course, the number of neurons may vary. For example, in practical applications, one or more specific indicators may be used to evaluate lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, target product yield, and inhibitor generation. For instance, glucose yield and xylose yield might be used simultaneously as specific indicators to evaluate the target product yield. Therefore, as the specific indicators increase, the corresponding number of neurons can be increased accordingly.

[0065] The hidden layers are set to 1-3 layers depending on the complexity of the target preprocessing technology. The network parameters are optimized using gradient descent to improve the accuracy and stability of the prediction model. By using the inherent property parameters of the lignocellulosic biomass powder samples described in S2 as samples, 70% of the samples are used as the training set and 30% as the validation set during model training to ensure the model's generalization ability.

[0066] The input layer uses min-max normalization to normalize the intrinsic property parameters of the lignocellulosic biomass powder samples in the training set, mapping them to the [0,1] interval; the activation function of the hidden layer is the modified linear unit (ReLU) function or the leaky modified linear unit (Leaky ReLU) function; the output layer uses a linear activation function to directly output the quantitative prediction results.

[0067] During network training, the weights were initialized using the He normal initialization method, with the initial value of the bias term set to 0.01. An adaptive moment estimation optimizer (Adam) was used for training, with mean squared error (MSE) as the loss function. A combined regularization strategy of L2 regularization and Dropout layers was employed to prevent overfitting. The samples were randomly divided into training and validation sets in a 7:3 ratio, and early stopping was used to control the number of training epochs. After training and optimization, the accuracy of the prediction model was significantly improved.

[0068] Specifically:

[0069] (1) The network is a feedforward fully connected network, with the following structure:

[0070] Input layer: This layer has 6 neurons, corresponding to cellulose content, hemicellulose content, lignin content, cellulose crystallinity, specific surface area, and porosity, respectively. The input layer performs min-max scaling on the intrinsic property parameters of the lignocellulosic biomass powder samples in the training set, mapping these parameters to the [0,1] interval to obtain normalized data. The formula is:

[0071]

[0072] This is to eliminate the interference of differences in the units of different parameters on network training. Among them, To train the intrinsic property parameters of concentrated lignocellulose biomass powder samples, This represents the minimum intrinsic property parameter of the lignocellulosic biomass powder samples in the training set. To train a set of intrinsic property parameters of lignocellulose biomass powder samples, The data is normalized. Furthermore, the input layer is only responsible for receiving and preprocessing the inherent property parameters of the lignocellulosic biomass powder samples in the training set; it does not perform any calculations, has no activation function, and directly passes the normalized data unidirectionally to the next hidden layer.

[0073] Hidden Layers: 1-3 layers can be flexibly configured according to the complexity of the target pretreatment process. These are the core computational layers of the network, responsible for uncovering the nonlinear correlation between the inherent properties and catalytic performance of lignocellulosic biomass raw materials. This invention preferably uses a two-layer hidden layer (denoted as the first hidden layer and the second hidden layer) as its basic architecture, adaptable to the prediction needs of most lignocellulosic biomass pretreatment processes.

[0074] Specifically, based on the characteristics of the alkaline hydrogen peroxide catalytic pretreatment process, this invention sets up two hidden layers. The first hidden layer is a feature extraction layer with 24 neurons, which performs high-dimensional mapping on the low-dimensional features in the normalized data transmitted from the input layer to extract high-dimensional features. The second hidden layer is a feature fusion layer with 12 neurons, which filters and fuses the high-dimensional features extracted by the first hidden layer, and outputs a feature vector that matches the output layer. Each hidden layer neuron consists of a weighted summation module and an activation function module. The activation function adopts the Modified Linear Unit (ReLU) function, with the following formula:

[0075]

[0076] To address the vanishing gradient problem of the traditional sigmoid function and improve network training efficiency and convergence speed, a Dropout layer is added to the output of each hidden layer, with a neuron dropout rate set to 0.2. This randomly masks some neuron connections to prevent overfitting.

[0077] Neuron structure in hidden layers: Each neuron in a hidden layer consists of a weighted summation module and an activation function module. First, the input from the upper layer is weighted and summed, and a bias term is added. Then, nonlinear characteristics are introduced through the activation function.

[0078] Inter-layer connections in hidden layers: The first hidden layer is fully connected to the input layer, and the second hidden layer is fully connected to the first hidden layer. Each neuron receives the output signals of all neurons in the previous layer.

[0079] Finally, if only one hidden layer is set in this invention, the number of neurons is configured to be 32-64, preferably 48.

[0080] Output Layer: The output layer of this invention contains 5 neurons, corresponding to lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, glucose yield, and phenolic inhibitor production, respectively. The output layer uses a linear activation function (Identity), with the following formula: This ensures that the output values ​​are consistent with the actual numerical range of the catalytic performance indicators, directly outputting quantitative prediction results. Specifically:

[0081] Number of neurons: fixed at 5, perfectly matching the number of key performance indicators for the catalytic reaction of lignocellulose biomass raw materials, with 1 neuron corresponding to 1 performance indicator.

[0082] The output variables corresponding to the neurons are, in order: lignin removal rate (%), cellulose conversion rate (%), hemicellulose conversion rate (%), target product yield (%), and inhibitor production (mg / g); if there are multiple target products / inhibitors, they are fused into a single dimension according to weights and then input into the corresponding neurons.

[0083] Activation function: A linear activation function (Identity) is used, with the following formula:

[0084]

[0085] The output values ​​are consistent with the actual numerical range of the catalytic performance indicators, eliminating the need for additional reverse normalization and directly outputting quantitative prediction results.

[0086] Layer function: The feature vectors output by the hidden layer are linearly weighted and calculated to output the final predicted values ​​of 5 catalytic performance indicators, providing a quantitative basis for screening raw materials for lignocellulosic biomass pretreatment.

[0087] (2) Network core parameters and initialization rules

[0088] Weights and biases

[0089] Weight matrix: The He normal initialization method is used to make the weight values ​​follow a mean of 0 and a variance of 0. The normal distribution; where This initialization method is adapted to the training characteristics of the ReLU activation function, representing the number of neurons in the upper layer.

[0090] Bias terms: All bias terms are initialized to 0.01 to avoid network training stagnation caused by the initial output of neurons being 0.

[0091] Network hyperparameters

[0092] Learning rate: A dynamic adaptive learning rate strategy is adopted, with the initial learning rate set to 0.001. During training, when the MSE loss value on the validation set does not decrease for 5 consecutive training epochs, the learning rate is multiplied by a decay coefficient of 0.5, and the minimum learning rate is limited to 0.00001 to prevent slow training convergence caused by an excessively small learning rate.

[0093] Batch size: Set to 16-32, with 24 being preferred in this invention, to balance network training efficiency and model generalization ability.

[0094] Training epochs: The maximum number of training epochs is set to 200, and early stopping is used. When the validation set loss value does not decrease for 10 consecutive training epochs, network training is stopped immediately to avoid model overfitting.

[0095] (3) Training and optimization mechanism

[0096] loss function

[0097] The mean squared error (MSE) function is used as the loss function, and the formula is as follows:

[0098]

[0099] in For the sample size, These are the actual measured values ​​of the catalytic performance indicators. This function, which outputs the predicted value from the network, can accurately measure the error between the predicted and actual values ​​and is suitable for regression prediction tasks of catalytic performance indicators.

[0100] Optimization Algorithm

[0101] An Adaptive Moment Estimation (Adam) optimizer is used instead of the basic gradient descent method to achieve adaptive adjustment of the learning rate, resulting in faster network convergence and stronger training stability. The core parameter of the Adam optimizer is set as: the exponential decay rate of the first-order moment estimate. ,

[0102] Second moment estimates the exponential decay rate. Numerical stability term .

[0103] Regularization strategy

[0104] A combined regularization strategy of L2 regularization and Dropout layers is adopted to effectively suppress model overfitting and improve model generalization ability. The weight decay coefficient of L2 regularization is set to 0.0001. Dropout layers are added to the output of hidden layer 1 and hidden layer 2 respectively. The neuron dropout rate of the Dropout layer is set to 0.2, and the connection relationship of some neurons is randomly masked.

[0105] Sample partitioning and training process

[0106] Sample splitting: The lignocellulosic biomass raw material samples were randomly divided into training set and validation set in a 7:3 ratio. The training set was used for iterative updates of network weights and bias terms, and the validation set was used for real-time monitoring of the model's generalization ability.

[0107] Training process: Iteratively train the network according to the following steps until the network converges, specifically: Min-Max normalization of the inherent property parameters of lignocellulosic biomass powder samples in the input layer training set → Feature extraction of hidden layer 1 → Random masking of dropout layer → Feature fusion of hidden layer 2 → Random masking of dropout layer → Prediction of catalytic performance index of output layer → Calculation of MSE loss value → Backpropagation of Adam optimizer → Update network weights and bias terms → Repeat the above steps.

[0108] The multilayer perceptron neural network structure of this invention is a modular and configurable architecture, which can be precisely adapted and adjusted according to the characteristics of different target preprocessing processes. The core adjustment rules are as follows: parameters not explicitly defined under each process adopt the default parameters described above in this invention:

[0109]

[0110] After the above training and optimization, the accuracy of the raw material performance prediction model can be greatly improved.

[0111] S6 The lignocellulosic biomass raw material to be screened is processed and tested according to the methods described in S1 to S2 to obtain its inherent property parameters; the inherent property parameters are input into the raw material performance prediction model constructed in S5, and the model outputs the performance index prediction value of the lignocellulosic biomass raw material to be screened under the target pretreatment process; a screening threshold is set, and if the performance index prediction value meets the screening threshold, the raw material is determined to be a suitable raw material; if it does not meet the threshold, it is determined to be an unsuitable raw material.

[0112] In this step, the screening threshold is set according to the performance requirements of the target pretreatment process, including the minimum values ​​for lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, target product yield, and inhibitor formation. The specific value of the screening threshold can be adjusted according to the quality standards and cost requirements of actual industrial production.

[0113] Example 1: Screening of lignocellulosic biomass feedstock adapted to alkaline hydrogen peroxide catalytic pretreatment process

[0114] S1 Select three types of lignocellulose biomass raw materials for screening: poplar, birch, and pine. Remove impurities such as bark and stones from each material, and pulverize them using a high-speed universal pulverizer. Then, sieve them through an 80-mesh standard inspection sieve to obtain lignocellulose biomass powder samples with uniform particle size. Place the three types of lignocellulose biomass powder samples into a vacuum drying oven and dry them at 70℃ for 3 hours. After drying, the moisture content of each sample is controlled at 5-7 wt% for later use.

[0115] S2 used the paradoxical separation method to determine the cellulose and hemicellulose content of the three samples, the Klason method to determine the lignin content, X-ray diffraction to determine the cellulose crystallinity, and nitrogen adsorption-desorption to determine the specific surface area and porosity. The results are shown in Table 1 below.

[0116] Table 1. Inherent property parameters of three lignocellulosic biomass feedstocks

[0117]

[0118] The S3 target pretreatment process is alkaline hydrogen peroxide catalytic pretreatment. The corresponding rapid catalytic evaluation conditions are set as follows: catalytic reaction temperature 80℃ (consistent with the industrial process), catalytic reaction time 30min (the actual industrial reaction time is 300min, i.e., 1 / 10), catalyst is hydrogen peroxide (8% of the raw material mass), reaction medium is 5% sodium hydroxide solution, solid-liquid ratio is 1:15 (g / mL), and a rapid catalytic detection and evaluation system is constructed.

[0119] S4. Take 0.5g of each of the three prepared lignocellulose biomass powder samples and place them into the above rapid catalytic detection and evaluation system. Catalytic reaction is carried out using a constant temperature water bath shaker at a stirring speed of 150 r / min. After the reaction, centrifuge at 4000 r / min for 8 min to separate the solid and liquid products. High-performance liquid chromatography (HPLC) is used to determine the cellulose conversion rate, hemicellulose conversion rate, and the yields of glucose and xylose. Ultraviolet spectrophotometry is used to determine the lignin removal rate and the amount of phenolic inhibitors generated. The detection results are shown in Table 2 below.

[0120] Table 2 Key performance indicators of three types of lignocellulosic biomass feedstock products

[0121]

[0122] S5 uses the six inherent property parameters in Table 1 as input variables and the six key performance indicators in Table 2 as output variables. It selects 100 sets of sample data from different lignocellulosic biomass raw materials (covering various lignocellulosic biomass such as poplar, birch, pine, and willow). A multilayer perceptron artificial neural network is used to construct a performance prediction model. The input layer has 6 neurons, the output layer has 6 neurons (glucose yield and xylose yield are counted as independent output neurons, hence a total of 6 output neurons), and there are two hidden layers: the first layer has 24 neurons, and the second layer has 12 neurons. The activation function is ReLU, and the dropout rate is 0.2. The training set:training set:validation set ratio is 7:3. The Adam optimizer (initial learning rate 0.001) is used, and early stopping is used to control training. The final prediction model has high accuracy on the validation set, and the correlation coefficient R between the predicted values ​​of the key performance indicators and the measured values ​​of the rapid catalytic evaluation is high. 2 ≥0.95.

[0123] S6 sets the screening thresholds for the alkaline hydrogen peroxide catalytic pretreatment process: lignin removal rate ≥60%, cellulose conversion rate ≥45%, hemicellulose conversion rate ≥55%, glucose yield ≥42%, xylose yield ≥52%, and phenolic inhibitor generation ≤1.8mg / g. The inherent property parameters of the three raw materials to be screened (see Table 1) are input into the above prediction model. The prediction results are compared with the screening thresholds. It can be seen that the predicted performance indicators of poplar and birch meet the screening thresholds and are judged to be suitable raw materials. The predicted lignin removal rate (about 45%), cellulose conversion rate (about 33%), and target product yield of pine are all lower than the screening thresholds, and the phenolic inhibitor generation exceeds the standard, so it is judged to be an unsuitable raw material. Verification experiment: Poplar, birch, and pine were pretreated using an actual industrial alkaline hydrogen peroxide catalytic pretreatment process (reaction time 300 min, other conditions consistent with the rapid catalytic evaluation system). Actual performance indicators were measured. The results showed that the actual performance indicators of poplar and birch deviated from the predicted indicators by ≤5%, indicating high pretreatment efficiency and stable product quality. The actual performance indicators of pine were consistent with the predicted indicators, but the pretreatment efficiency was low, further demonstrating the accuracy and reliability of the screening method of this invention.

[0124] Example 2: Screening of lignocellulosic biomass feedstock adapted to solid acid catalytic pretreatment process

[0125] S1 selected three types of lignocellulose biomass raw materials for screening: willow, maple, and fir. After removing impurities, the raw materials were pulverized using a high-speed universal pulverizer and sieved through a 60-mesh standard sieve to obtain lignocellulose biomass powder samples. The powder samples were then dried in a vacuum drying oven at 65℃ for 4 hours, with a moisture content of ≤8wt% after drying, and were ready for use.

[0126] S2 uses the same detection method as in Example 1 to determine the inherent property parameters of the three raw materials; specific data are omitted.

[0127] The S3 target pretreatment process is solid acid catalytic pretreatment. The rapid catalytic evaluation conditions are set as follows: catalytic reaction temperature 90℃, catalytic reaction time 40min (actual industrial reaction time is 200min, i.e. 1 / 5), catalyst is sulfonated biochar (5% of the raw material mass), reaction medium is deionized water, and solid-liquid ratio is 1:12 (g / mL). A rapid catalytic evaluation system is constructed.

[0128] S4 Take a spare lignocellulose biomass powder sample and put it into the above rapid catalytic evaluation system. Carry out the catalytic reaction in a high-pressure reactor with a stirring speed of 180 r / min. After the reaction is completed, centrifuge to separate the product and determine the key performance indicators using the same method as in Example 1. Specific data are omitted.

[0129] S5 uses the same machine learning algorithm as Example 1, with the inherent property parameters of the raw materials as input variables, and the same multilayer perceptron neural network algorithm as Example 1, with the key performance indicators of the catalytic reaction as output variables, to construct a performance prediction model. The model has a high accuracy.

[0130] S6 sets the screening threshold for the solid acid catalytic pretreatment process. The inherent property parameters of the three raw materials are input into the prediction model to predict performance indicators, which are then compared with the threshold. The results show that the predicted performance indicators of willow and maple meet the screening threshold, and they are determined to be suitable raw materials. The predicted performance indicators of fir do not meet the screening threshold (high lignin content, low predicted removal efficiency), and they are determined to be unsuitable raw materials. Verification through actual industrial solid acid catalytic pretreatment process shows that the screening results are accurate and reliable, with a deviation ≤6%.

[0131] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A rapid catalytic evaluation method for the properties of lignocellulose biomass, characterized in that, Includes the following steps: S1 Select lignocellulose biomass raw material, remove impurities, crush and sieve to obtain lignocellulose biomass powder sample with uniform particle size; dry the lignocellulose biomass powder sample to remove moisture and set aside for later use. S2 Detects the inherent property parameters of the lignocellulose biomass powder sample, including: cellulose content, hemicellulose content, lignin content, cellulose crystallinity, specific surface area, and porosity; S3. Based on the target pretreatment process, determine the corresponding rapid catalytic evaluation conditions to construct a rapid catalytic detection and evaluation system; the rapid catalytic evaluation conditions include: catalytic reaction temperature, catalytic reaction time, catalyst type and dosage, reaction medium and concentration, and solid-liquid ratio; the catalytic reaction time is 1 / 10 to 1 / 5 of the actual industrial pretreatment reaction time; S4 Take the lignocellulose biomass powder sample and place it into the rapid catalytic detection and evaluation system of S3 for catalytic reaction; after the reaction is completed, separate the reaction products and detect the key performance indicators of the reaction products; the key performance indicators include: lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, target product yield and inhibitor generation amount; S5 uses the inherent property parameters of the lignocellulosic biomass powder sample detected in S2 as input variables and the key performance indicators detected in S4 as output variables to construct a raw material performance prediction model using a multilayer perceptron artificial neural network. S6 The lignocellulosic biomass raw material to be screened is processed and tested according to the methods described in S1 to S2 to obtain its inherent property parameters; the inherent property parameters are input into the raw material performance prediction model constructed in S5, and the model outputs the performance index prediction value of the lignocellulosic biomass raw material to be screened under the target pretreatment process; a screening threshold is set, and if the performance index prediction value meets the screening threshold, the raw material is determined to be a suitable raw material; if it does not meet the threshold, it is determined to be an unsuitable raw material.

2. The method according to claim 1, characterized in that, In S3, the target pretreatment process includes one of the following: alkaline hydrogen peroxide catalytic pretreatment, solid acid catalytic pretreatment, ionic liquid synergistic catalytic pretreatment, and mechanical-enzyme synergistic catalytic pretreatment.

3. The method according to claim 1, characterized in that, In S1, the pulverization is carried out using a high-speed universal pulverizer, and the sieving is carried out using a standard inspection sieve. The particle size of the lignocellulose biomass powder sample is 40-100 mesh. The drying is carried out using a vacuum drying oven at a temperature of 60-80℃ for 2-4 hours. The moisture content after drying is ≤8wt%.

4. The method according to claim 1, characterized in that, In S2, the cellulose and hemicellulose contents were determined by the Paraná separation method, the lignin content was determined by the Klason method, the cellulose crystallinity was determined by X-ray diffraction, and the specific surface area and porosity were determined by nitrogen adsorption-desorption method.

5. The method according to claim 1, characterized in that, In S4, the catalytic reaction is carried out using a constant temperature water bath shaker or a high-pressure reactor, and the stirring speed is controlled at 100-200 r / min during the reaction. The reaction products are separated by centrifugation, with a centrifugation speed of 3000-5000 r / min and a centrifugation time of 5-10 min.

6. The method according to claim 1, characterized in that, In S5, the multilayer perceptron artificial neural network is a feedforward fully connected network, including an input layer, a hidden layer, and an output layer; the input layer has neurons corresponding to the inherent property parameters of the lignocellulosic biomass powder sample; the output layer has neurons corresponding to the key performance indicators; and the hidden layer is set to 1-3 layers according to the target pretreatment process.

7. The method according to claim 6, characterized in that, The input layer uses min-max normalization to normalize the inherent property parameters of the lignocellulosic biomass powder samples in the training set, mapping them to the [0,1] interval; the activation function of the hidden layer uses a modified linear unit function or a modified linear unit function with leakage; the output layer uses a linear activation function to directly output the quantitative prediction results.

8. The method according to claim 7, characterized in that, The weights were initialized using the He normal initialization method, with the initial value of the bias term set to 0.

01. An adaptive moment estimation optimizer was used for training, with mean squared error as the loss function. A combined regularization strategy of L2 regularization and Dropout layer was used to prevent overfitting. The samples were randomly divided into training and validation sets in a 7:3 ratio, and the early stopping method was used to control the number of training rounds.

9. The method according to claim 1, characterized in that, The screening thresholds described in S6 are set according to the performance requirements of the target pretreatment process, including the minimum values ​​of lignin removal rate, cellulose conversion rate, hemicellulose conversion rate, target product yield, and inhibitor generation.

10. The method according to claim 1, characterized in that, The catalyst includes one or more of solid acid catalysts, metal complex catalysts, and enzyme catalysts; the reaction medium includes one or more of deionized water, dilute acid solution, dilute alkali solution, or ionic liquid, with a solid-liquid ratio of 1:10 g / mL to 1:20 g / mL.

Images