Multi-modal data fused pine pollen airflow wall breaking parameter optimization method and system

Abstract

The invention relates to a multi-modal data fused pine pollen airflow wall breaking parameter optimization method and system, and relates to the technical field of wall breaking parameter optimization, and the method comprises the steps: in an airflow wall breaking parameter adjustment space, taking maximization of wall breaking quality and minimization of equipment energy consumption as double-target optimization, and determining an initial optimal airflow wall breaking parameter; controlling the jet mill to perform wall breaking treatment on the to-be-processed pine pollen, and setting an adaptive adjustment time window based on the actual wall breaking control parameter sequence; optimizing and adjusting the initial optimal airflow wall breaking parameters according to the multi-source wall breaking monitoring data sequence, and outputting the current optimal airflow wall breaking parameters; and updating the control instruction according to the current optimal airflow wall breaking parameter, continuing to execute wall breaking treatment and performing closed-loop iteration regulation and control until the wall breaking treatment of all the to-be-processed pine pollen is completed. The problems that traditional pine pollen airflow wall breaking depends on fixed parameters, adaptability is poor, and wall breaking quality and energy consumption are difficult to balance are solved.

Description

Technical Field

[0001] This application relates to the field of cell wall disruption parameter optimization, and in particular to a method and system for optimizing pine pollen airflow cell wall disruption parameters through multimodal data fusion. Background Technology

[0002] With the expanding application of pine pollen in health products and pharmaceuticals, the quality stability and energy efficiency of its airflow cell-wall breaking processing have become key technological requirements for industrial production.

[0003] Currently, traditional airflow cell-wall breaking methods for pine pollen rely on fixed parameters for processing. This not only fails to dynamically adapt the parameters according to the actual material state of the pine pollen to be processed, resulting in poor parameter adaptability, but also makes it difficult to balance cell-wall breaking quality and equipment energy consumption during processing. This not only easily causes fluctuations in cell-wall breaking effect, but also increases the waste of production resources and reduces the overall efficiency of pine pollen processing. Summary of the Invention

[0004] This application provides a method and system for optimizing the parameters of pine pollen airflow cell disruption by multimodal data fusion, which improves the current situation where traditional pine pollen airflow cell disruption relies on fixed parameters, has poor adaptability, and is difficult to balance cell disruption quality and equipment energy consumption during processing.

[0005] The embodiments of this application disclose the following technical solutions: In a first aspect, embodiments of this application provide a method for optimizing pine pollen airflow cell wall disruption parameters through multimodal data fusion, the method comprising: Based on the material state data of the pine pollen to be processed, within the adjustment space of the airflow cell disruption parameters, the initial optimal airflow cell disruption parameters are determined by optimizing the dual objectives of maximizing cell disruption quality and minimizing equipment energy consumption. The airflow pulverizer is controlled to break the pine pollen to be processed according to the initial optimal airflow cell breaking parameters. During the cell breaking process, an adaptation adjustment time window is set for a preset future period based on the actual cell breaking control parameter sequence obtained by real-time monitoring. Within the adaptation and adjustment time window, the initial optimal airflow cell disruption parameters are optimized and adjusted according to the multi-source cell disruption monitoring data sequence composed of acoustic signals and machine vision data, and the current optimal airflow cell disruption parameters are output. The control command is updated with the current optimal airflow cell-breaking parameters to continue the cell-breaking process. Based on the adaptive adjustment time window, the cell-breaking process and monitoring steps are dynamically returned to perform closed-loop iterative regulation until all the pine pollen to be processed has been cell-breaking.

[0006] Secondly, embodiments of this application provide a multimodal data fusion-based pine pollen airflow cell disruption parameter optimization system, the system comprising: The initial parameter optimization module is used to optimize the initial optimal airflow cell disruption parameters based on the material state data of the pine pollen to be processed, within the adjustment space of the airflow cell disruption parameters, with the dual objectives of maximizing cell disruption quality and minimizing equipment energy consumption. The adaptation window setting module is used to control the airflow pulverizer to break the pine pollen to be processed according to the initial optimal airflow cell breaking parameters, and to set an adaptation adjustment time window in the future period based on the actual cell breaking control parameter sequence obtained by real-time monitoring during the cell breaking process. The multi-source data adjustment module is used to optimize and adjust the initial optimal airflow cell breaking parameters according to the multi-source cell breaking monitoring data sequence composed of acoustic signals and machine vision data within the adaptation adjustment time window, and output the current optimal airflow cell breaking parameters. The closed-loop iterative control module is used to update the control command with the current optimal airflow cell-breaking parameters, continue to execute the cell-breaking process, and dynamically return to the cell-breaking process and monitoring steps based on the adaptive adjustment time window to perform closed-loop iterative control until all the pine pollen to be processed has been cell-breaking.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application proposes a method and system for optimizing parameters of pine pollen airflow cell disruption using multimodal data fusion. By carrying out parameter optimization before processing, setting an appropriate time window, multi-source data regulation and closed-loop iterative control in steps, the optimization and control of parameters during the pine pollen airflow cell disruption process are realized. First, material state data of the pine pollen to be processed is collected, and the adjustment space of the airflow pulverizer's cell-wall breaking parameters is simultaneously obtained. Then, based on the material state data and parameter adjustment space, combined with the pine pollen cell-wall breaking quality predictor and the airflow pulverizer twin simulation model, the initial optimal airflow cell-wall breaking parameters are determined through a genetic learning algorithm. Subsequently, the equipment is controlled to start the cell-wall breaking process according to the initial optimal airflow cell-wall breaking parameters. During the process, the actual control parameter sequence within a preset historical period is monitored, the overall control deviation and control fluctuation coefficient are calculated, and the time window compensation coefficient is obtained by combining the current continuous processing time. An adaptation adjustment time window is set within a preset future period. Next, within the adaptation adjustment time window, acoustic signals and machine vision data are collected. After feature extraction, the data is input into the control error prediction plugin to predict the control error. Based on this, the initial optimal airflow cell-wall breaking parameters are fine-tuned to obtain the current optimal airflow cell-wall breaking parameters. Finally, the control command is updated with the current optimal airflow cell-wall breaking parameters. The cell-wall breaking process and monitoring steps are dynamically returned based on the adaptation adjustment time window, and the closed-loop iteration continues until all the pine pollen to be processed is completed.

[0008] The technical solution of this application solves the problems of poor adaptability and difficulty in balancing the quality of cell disruption and energy consumption caused by the reliance on fixed parameters in traditional pine pollen airflow cell disruption. It avoids insufficient cell disruption, loss of heat-sensitive nutrients, or excessive energy consumption of equipment caused by fixed parameters. At the same time, by dynamically setting the time window for adaptation and adjustment, it ensures that the frequency of parameter fine-tuning is accurately matched with the actual processing needs, further improving the stability and economy of pine pollen airflow cell disruption. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0010] Figure 1 A flowchart illustrating the method for optimizing pine pollen airflow cell disruption parameters through multimodal data fusion, as provided in this application embodiment;

[0011] Figure 2 This is a schematic diagram of the structure of the pine pollen airflow cell wall breaking parameter optimization system for multimodal data fusion provided in the embodiments of this application.

[0012] The components represented by each number in the attached diagram are explained below: Initial parameter optimization module 01, adaptation window setting module 02, multi-source data adjustment module 03, closed-loop iterative control module 04. Detailed Implementation

[0013] This application provides a method and system for optimizing the parameters of pine pollen airflow cell disruption using multimodal data fusion. This method addresses the technical problems in existing technologies, such as the reliance on fixed parameters for traditional pine pollen airflow cell disruption, poor adaptability, and difficulty in balancing cell disruption quality and equipment energy consumption during processing, which leads to unstable cell disruption effects and wasted production resources.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0016] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0017] Example 1, as shown in the appendix Figure 1 As shown, this application provides a method for optimizing pine pollen airflow cell wall disruption parameters through multimodal data fusion, the method comprising the following steps: S110: Based on the material state data of the pine pollen to be processed, within the adjustment space of the airflow cell disruption parameters, optimization is performed with the dual objectives of maximizing cell disruption quality and minimizing equipment energy consumption to determine the initial optimal airflow cell disruption parameters; In this embodiment of the application, in the scenario of airflow cell wall breaking processing of pine pollen, in order to avoid the problems of poor adaptability and difficulty in balancing cell wall breaking quality and energy consumption caused by traditional fixed parameter processing, it is necessary to first obtain the material characteristics of the pine pollen to be processed, clarify the adjustable parameter range of the airflow pulverizer, and then determine the initial optimal airflow cell wall breaking parameters through multi-step collaborative optimization to ensure that the subsequent cell wall breaking processing can match the material characteristics.

[0018] Specifically, for the pine pollen to be processed, the initial particle size distribution, toughness, hardness and moisture content are collected and obtained. These characteristic data that directly affect the cell wall breaking effect are used as material state data to provide a basis for subsequent parameter adaptation.

[0019] Meanwhile, the adjustment range of the airflow cell disruption parameters for the airflow mill is clearly defined. This adjustment range encompasses four adjustable parameters: airflow pressure, feed rate, milling chamber temperature, and classifier speed, thus defining the boundaries of parameter optimization.

[0020] Furthermore, within the defined adjustment space of airflow cell disruption parameters, several initial airflow cell disruption parameters are randomly selected, and each initial airflow cell disruption parameter is combined with the aforementioned material state data to form several targeted pine pollen cell disruption schemes, so as to ensure that each scheme can be associated with specific material characteristics and processing parameters.

[0021] Furthermore, using a pre-constructed pine pollen cell wall breaking quality predictor and a twin simulation model of an airflow pulverizer, each pine pollen cell wall breaking scheme was simulated and analyzed. Based on the predictor, the cell wall breaking quality coefficient of the corresponding scheme was obtained, and based on the twin simulation model, the simulated cell wall breaking energy consumption of the corresponding scheme was obtained, providing a data basis for scheme performance evaluation.

[0022] Furthermore, with the dual objectives of maximizing cell wall disruption quality and minimizing equipment energy consumption, a parameter fitness evaluation function is constructed. The cell wall disruption quality coefficient and simulated energy consumption of each pine pollen cell wall disruption scheme are substituted into the parameter fitness evaluation function to calculate the parameter fitness corresponding to each scheme. The comprehensive performance of the scheme is quantified by the fitness value.

[0023] Finally, using a genetic learning algorithm, with several initial airflow wall-breaking parameters as the optimization object and the corresponding parameter fitness as the evaluation criterion, iterative optimization is performed to select the parameter combination with the highest fitness and output the initial optimal airflow wall-breaking parameters.

[0024] This step establishes a foundation by first collecting material state data and defining the adjustment space for airflow cell disruption parameters. Then, through a process of scheme generation, performance prediction, fitness calculation, and algorithm optimization, the initial optimal airflow cell disruption parameters that are suitable for material characteristics and balance quality and energy consumption are determined, providing a parameter basis for the effective control and adjustment of subsequent cell disruption processing.

[0025] Step S110 in the method provided in this application embodiment includes: The initial particle size distribution, toughness, hardness and moisture content of the pine pollen to be processed were collected and obtained as material state data. The adjustment range of airflow cell breaking parameters for the airflow pulverizer is obtained, wherein the airflow cell breaking parameters include airflow pressure, feed rate, pulverizing chamber temperature and classifying wheel speed; Within the airflow cell disruption parameter adjustment space, several initial airflow cell disruption parameters are randomly selected and combined with the material state data to generate several pine pollen cell disruption schemes. Using a pre-built pine pollen cell wall breaking quality predictor and a twin simulation model of the airflow pulverizer, several cell wall breaking quality coefficients and several cell wall breaking simulation energy consumptions are predicted based on several pine pollen cell wall breaking schemes. A parameter fitness evaluation function is constructed with the dual objectives of maximizing cell wall breaking quality and minimizing equipment energy consumption. Several parameter fitness values ​​are calculated based on the aforementioned cell wall breaking quality coefficients and several cell wall breaking simulation energy consumption values. Using a genetic learning algorithm, optimization is performed based on several initial airflow wall-breaking parameters and several parameter fitnesss to output the initial optimal airflow wall-breaking parameters.

[0026] In this embodiment of the application, in order to avoid the problems of unstable cell wall breaking quality and high equipment energy consumption caused by poor parameter adaptability of pine pollen airflow cell wall breaking, it is necessary to first clarify the material characteristics of the pine pollen to be processed and the adjustable parameter range of the airflow pulverizer, and then determine the initial optimal airflow cell wall breaking parameters that can balance cell wall breaking quality and energy consumption through a multi-stage collaborative parameter optimization process, so as to ensure the effect of subsequent cell wall breaking processing.

[0027] Specifically, the initial particle size distribution, toughness, hardness, and moisture content of the pine pollen to be processed are collected and obtained to understand its material characteristics. The initial particle size distribution can be obtained using a laser particle size analyzer, the toughness using a texture analyzer, the hardness using a microhardness tester, and the moisture content using a rapid moisture analyzer.

[0028] Furthermore, the adjustment space of the airflow cell disruption parameters of the airflow pulverizer is obtained to clarify the controllable range of parameters and provide a reasonable basis for subsequent random selection of initial parameters and generation of cell disruption schemes.

[0029] The adjustable parameters for airflow pulverization include airflow pressure, feed rate, pulverizing chamber temperature, and classifying wheel speed. These parameters can be obtained from the equipment operation manual of the airflow pulverizer. The manual will clearly indicate the adjustable range, safety threshold, and basic setting standards for each parameter.

[0030] Furthermore, within the determined airflow cell disruption parameter adjustment space, several initial airflow cell disruption parameters are randomly selected, and then each initial parameter is combined with the previously obtained material state data to generate several pine pollen cell disruption schemes.

[0031] For example, a combination of parameters such as airflow pressure of 1.2 MPa, feeding speed of 25 kg / h, crushing chamber temperature of 42°C, and classifying wheel speed of 3200 r / min is selected and combined with the corresponding material state data to form a unique pine pollen cell wall breaking solution, so as to ensure that each solution can associate specific material characteristics and processing parameters.

[0032] Furthermore, using a pre-built pine pollen cell wall breaking quality predictor and a twin simulation model of an airflow pulverizer, each pine pollen cell wall breaking scheme was simulated and analyzed, and the cell wall breaking quality coefficient of the corresponding scheme was obtained through the pine pollen cell wall breaking quality predictor.

[0033] The method provided in this application includes the following steps in constructing the pine pollen cell wall breakage quality predictor: Based on historical airflow cell wall breaking processing records of pine pollen, a sample cell wall breaking scheme set and a sample cell wall breaking quality dataset were collected. The sample cell wall breaking quality data were evaluated to determine the sample cell wall breaking quality coefficient and obtain a sample cell wall breaking quality coefficient set. The cell wall breaking quality data includes at least the cell wall breaking rate, particle size distribution and key nutrient retention rate.

[0034] Using the sample cell wall breaking scheme set as input, and the sample cell wall breaking quality dataset as supervision, a deep learning model is trained until convergence to generate a pine pollen cell wall breaking quality predictor.

[0035] Specifically, based on historical airflow cell disruption processing records of pine pollen, a sample cell disruption scheme set and a sample cell disruption quality dataset were first collected.

[0036] The sample cell disruption scheme set was selected from representative past processing cases, covering schemes with different material states and different parameter combinations to ensure the comprehensiveness of the samples. For example, schemes such as "pine pollen initial particle size 15-25μm, moisture content 7%, combined with air pressure 1.2MPa, feed rate 25kg / h" and "pine pollen initial particle size 20-30μm, moisture content 9%, combined with air pressure 1.5MPa, feed rate 20kg / h" were collected to form the sample cell disruption scheme set.

[0037] In addition, the sample cell wall disruption quality dataset is the test result after the corresponding sample cell wall disruption scheme is processed. The test result includes at least the cell wall disruption rate, particle size distribution and key nutrient retention rate.

[0038] Among these factors, the cell wall breakage rate reflects the proportion of cell wall fragmentation, particle size distribution affects subsequent absorption efficiency, and the retention rate of key nutrients affects the efficacy of the final product. For example, a sample cell wall breakage solution, after testing, showed a cell wall breakage rate of 92%, a particle size of 10-20μm of 85%, and a key nutrient retention rate of 88%. These data constitute the sample cell wall breakage quality data for that solution.

[0039] Furthermore, the cell wall disruption quality data of each sample are evaluated to determine the cell wall disruption quality coefficient, and the sample cell wall disruption quality coefficient set is obtained by integration.

[0040] Specifically, the evaluation should be conducted in conjunction with industry standards and product requirements. First, scoring criteria should be set for cell wall breakage rate, particle size distribution, and retention rate of key nutrients. Then, the cell wall breakage quality coefficient of the sample should be calculated by weighting according to preset weights, and a score of 0-100 should be obtained.

[0041] For example, a cell wall breakage rate of ≥90% earns 30 points, a cell wall breakage rate of 80%-89% earns 20 points, a particle size of 10-20μm accounting for ≥80% earns 30 points, a particle size of 70%-79% earns 20 points, a key nutrient retention rate of ≥85% earns 40 points, and a key nutrient retention rate of 75%-84% earns 30 points.

[0042] If a sample cell disruption scheme has a cell disruption rate of 92% (30 points), a particle size distribution of 85% (30 points), and a key nutrient retention rate of 88% (40 points), the sample cell disruption quality coefficient is 100 points; if another sample cell disruption scheme has a cell disruption rate of 85% (20 points), a particle size distribution of 75% (20 points), and a key nutrient retention rate of 80% (30 points), the sample cell disruption quality coefficient is 70 points. All coefficient results are summarized to form a sample cell disruption quality coefficient set.

[0043] Furthermore, using a sample set of cell wall disruption schemes as input and a sample set of cell wall disruption quality dataset as supervision, a deep learning model is trained until convergence, generating a pine pollen cell wall disruption quality predictor.

[0044] Specifically, considering the nonlinear correlation between the sample cell disruption scheme and the sample cell disruption quality coefficient, a pine pollen cell disruption quality predictor can be constructed based on a neural network model containing an input layer, a hidden layer, and an output layer.

[0045] The input layer receives parameter data from the sample cell disruption scheme set, the hidden layer analyzes the complex correlation between parameters using the ReLU function, and the output layer corresponds to the cell disruption quality coefficient of the sample, thereby realizing the mapping prediction from scheme to quality coefficient.

[0046] Before model training, the sample data needs to be preprocessed, that is, the structured parameters in the sample cell wall breaking scheme set are standardized. The Z-score standardization formula "standardized value = (original value - mean) / standard deviation" is used to convert each parameter into a distribution with a mean of 0 and a standard deviation of 1, thereby eliminating the interference of different parameter magnitude differences on model training.

[0047] Meanwhile, the sample cell wall breakage quality coefficient is normalized and mapped to the [0, 1] interval to ensure that the model output and the numerical range of the supervision signal are consistent, thereby improving training efficiency. For example, the airflow pressure of 1.2 MPa is normalized to 0.8, and the sample cell wall breakage quality coefficient of 100 is normalized to 1.0 to ensure that the data format meets the model input requirements.

[0048] Furthermore, the preprocessed sample cell disruption scheme set is used as the model input, and the sample cell disruption quality coefficient set is used as the supervision signal to carry out supervised training.

[0049] Specifically, the samples are first divided into a training set and a validation set in an 8:2 ratio. The training set is used for model parameter learning, and the validation set is used to evaluate the model's generalization ability. During training, the mean squared error loss function is used to calculate the deviation between the model's predicted cell wall breaking quality coefficient and the actual labeled coefficient. The model's weights and bias parameters are adjusted layer by layer through the backpropagation algorithm to continuously reduce the loss value.

[0050] After each iteration, the model performance is tested using a validation set, and the validation set loss value is recorded. If the validation set loss value does not decrease and tends to stabilize after 15 consecutive iterations, the model training is considered to have converged, and the iteration is stopped.

[0051] After training convergence, the model's performance needs to be validated. Historical cell wall breaking cases that were not used in training are selected as the test set. The sample cell wall breaking schemes in the test set are input into the model, and the cell wall breaking quality coefficient predicted by the model is compared with the actual detected quality coefficient. If the average absolute error of the prediction is ≤5%, it indicates that the model's prediction accuracy meets the requirements and can be determined as the final pine pollen cell wall breaking quality predictor.

[0052] Furthermore, using a twin simulation model of an airflow pulverizer, several simulated energy consumptions for cell wall breaking were predicted based on several pine pollen cell wall breaking schemes.

[0053] Among them, the twin simulation model of the airflow pulverizer is a digital model built based on a physics engine and historical operating data. It can reproduce the actual operating state of the airflow pulverizer in a virtual environment. The core is to simulate the energy consumption of the equipment under different pine pollen cell wall breaking schemes by mapping the structural parameters, working principle and energy consumption characteristics of the real equipment.

[0054] During the simulation and prediction, the operating parameters such as airflow pressure, feeding speed, and classifying wheel speed in the pine pollen cell wall breaking scheme need to be input into the twin simulation model. The model will calculate the compressed air energy consumption required for airflow generation, the motor energy consumption for classifying wheel rotation, and the load energy consumption of the equipment under different feeding speeds based on the parameters, and finally summarize and output the cell wall breaking simulation energy consumption of the scheme (unit: kW·h / ton of pine pollen).

[0055] For example, for the pine pollen cell wall breaking scheme with "airflow pressure of 1.2MPa and feeding speed of 25kg / h" mentioned above, the twin simulation model calculates that the compressed air energy consumption is 80kW·h / ton, the motor energy consumption is 45kW·h / ton, and the total simulated cell wall breaking energy consumption is 125kW·h / ton; if another scheme increases the airflow pressure to 1.5MPa and reduces the feeding speed to 20kg / h, the simulated cell wall breaking energy consumption output by the model is 142kW·h / ton.

[0056] Furthermore, a parameter fitness evaluation function is constructed with the dual objectives of maximizing cell wall disruption quality and minimizing equipment energy consumption, and the corresponding parameter fitness is calculated based on the obtained cell wall disruption quality coefficients and the energy consumption of multiple cell wall disruption simulations.

[0057] Since there is a certain contradiction between cell wall breaking quality and energy consumption (i.e., increasing airflow pressure may improve cell wall breaking quality, but will increase energy consumption), it is necessary to construct a parameter fitness evaluation function to balance the weights of the two, transforming the dual-objective optimization into a single-dimensional fitness value. The fitness value ranges from 0 to 1, with higher values ​​indicating better overall performance of the solution.

[0058] Specifically, the cell wall breaking quality coefficient and the cell wall breaking simulation energy consumption need to be normalized first. That is, the cell wall breaking quality coefficient (0-100 points) is transformed into a normalized quality value through linear mapping. The normalized quality value is denoted as Q. The specific calculation formula can be expressed as "Q=cell wall breaking quality coefficient / 100", so that the cell wall breaking quality performance is mapped to the interval [0,1].

[0059] In addition, the simulated energy consumption of the cell wall breaking process is reverse-normalized according to the reasonable energy consumption range of the airflow pulverizer. The normalized energy consumption value is denoted as E', and the specific calculation formula can be expressed as "E'=(E max -E) / (E max -E min "), where E is the simulated energy consumption for cell wall disruption, E max E is the upper limit of energy consumption. min As the lower limit of energy consumption, the lower the energy consumption of the cell disruption simulation, the closer E' is to 1, and the higher the energy consumption, the closer E' is to 0, so as to ensure that the optimization direction of energy consumption indicators and quality indicators is consistent.

[0060] Furthermore, a fitness evaluation function is constructed. First, a quality weight coefficient α and an energy consumption weight coefficient β are introduced, which must satisfy α+β=1. The specific fitness evaluation function formula is "fitness = α×Q+β×E'".

[0061] The weight values ​​need to be determined based on the actual needs of pine pollen processing. For example, when producing high value-added products, α should be 0.6 and β should be 0.4 to prioritize quality assurance; when producing in batches, α should be 0.4 and β should be 0.6 to prioritize energy consumption control.

[0062] For example, if α=0.5 and β=0.5, for the above-mentioned pine pollen cell wall breaking scheme with a cell wall breaking quality coefficient of 92 and a simulated energy consumption of 125 kW·h / ton: Q=92 / 100=0.92, assuming E max =180, E min=100, E'=(180-125) / (180-100)=0.6875, substituting into the fitness evaluation function, we get fitness=0.5×0.92+0.5×0.6875=0.80375.

[0063] In addition, for another pine pollen cell wall breaking scheme with a cell wall breaking quality coefficient of 95 and a simulated energy consumption of 142 kW·h / ton: Q=0.95, E'=(180-142) / 80=0.475, fitness=0.5×0.95+0.5×0.475=0.7125; it can be seen from the comparison that although the quality of the first scheme is slightly lower, it has lower energy consumption and higher overall fitness, and is more in line with the dual-objective optimization requirements.

[0064] Furthermore, by utilizing the genetic learning algorithm in the existing technology, optimization is performed based on multiple randomly selected initial airflow wall-breaking parameters and corresponding fitness values ​​to output the initial optimal airflow wall-breaking parameters.

[0065] Specifically, the initialization population step of the genetic learning algorithm is executed first. The airflow cell disruption parameters to be optimized need to be determined as decision variables, namely airflow pressure, feed rate, crushing chamber temperature, and classifier speed. These parameters are all derived from the airflow cell disruption parameter adjustment space.

[0066] Furthermore, the operating instructions for the air jet mill should specify the range of values ​​for each air jet cell breaking parameter, such as air pressure range of 0.5-2.0 MPa, feed rate range of 10-50 kg / h, grinding chamber temperature range of 30-60℃, and classifying wheel speed range of 2000-4000 r / min, to ensure the compliance and feasibility of parameter selection.

[0067] Furthermore, an initial population containing N individuals is randomly generated. Each individual is a set of airflow disruption parameters, which can be represented as a chromosome [airflow pressure, feed rate, crushing chamber temperature, classifier rotation speed].

[0068] For example, one individual is generated with [1.2MPa, 25kg / h, 40℃, 3000r / min], and another individual is generated with [1.5MPa, 20kg / h, 45℃, 3500r / min], thus forming an initial population that covers different possible combinations within the parameter adjustment space.

[0069] Furthermore, the fitness value of each individual is calculated. This involves combining each initial airflow cell disruption parameter combination with the corresponding material state data to generate a pine pollen cell disruption scheme. Then, the cell disruption quality coefficient and cell disruption simulation energy consumption are obtained through a pine pollen cell disruption quality predictor and a twin simulation model, respectively. These values ​​are then substituted into the constructed parameter fitness evaluation function to calculate the fitness value.

[0070] For example, a certain initial parameter combination [1.2MPa, 25kg / h, 40℃, 3000r / min] corresponds to a cell disruption quality coefficient of 92 points, and the simulated cell disruption energy consumption is 125kW·h / ton. If the mass weight coefficient α=0.5 and the energy consumption weight coefficient β=0.5, the fitness value is 0.80375. Another individual [1.5MPa, 20kg / h, 45℃, 3500r / min] corresponds to a cell disruption quality coefficient of 95 points, and the simulated cell disruption energy consumption is 142kW·h / ton, with a fitness value of 0.7125. By assigning a fitness score to each individual in the above manner, the overall performance of the individual can be reflected.

[0071] Next, a selection process is performed. Following the roulette wheel selection method, parent individuals for reproduction are chosen from the current population. The core principle is that individuals with higher fitness values ​​have a greater probability of being selected, thus simulating natural selection and the process of survival of the fittest.

[0072] For example, the fitness values ​​of all individuals in the population are summed, and the proportion of each individual's fitness value to the total is calculated. This proportion is the probability of the individual being selected. If an individual's fitness value is 0.80375, accounting for 25%, then its probability of being selected is 25%. Through the probabilistic selection process described above, individuals with high fitness have a greater chance of participating in reproduction and passing on their superior traits to their offspring.

[0073] Next, a crossover operation is performed. The parent individuals selected in the previous step are paired up, and with a certain crossover probability, some genes (i.e., airflow cell wall breaking parameters) on the paired chromosomes are exchanged to produce new offspring individuals.

[0074] For example, assuming a crossover probability of 0.8, two parent individuals are [1.2MPa, 25kg / h, 40℃, 3000r / min] and [1.5MPa, 20kg / h, 45℃, 3500r / min], respectively. After crossover at a random site, offspring [1.2MPa, 20kg / h, 45℃, 3000r / min] and [1.5MPa, 25kg / h, 40℃, 3500r / min] may be produced. The above crossover operation helps to combine the excellent parameter characteristics of the parents and expand the parameter diversity of the population.

[0075] Furthermore, a mutation operation is performed. With a low mutation probability, such as 0.1, the value of a certain gene in the offspring is randomly changed. For example, the feed rate of an offspring [1.2 MPa, 20 kg / h, 45℃, 3000 r / min] might be 20 kg / h, but could randomly vary within a small range to 21 kg / h. This mutation operation introduces new possible parameter combinations, helping to maintain population diversity and preventing the algorithm from prematurely falling into local optima.

[0076] Furthermore, a new population is formed and iterated. The new individuals generated through selection, crossover, and mutation are combined to form a new generation of population. The steps of calculating fitness, selection, crossover, and mutation are repeated to iteratively optimize the population. Typically, a maximum number of iterations is set, or the algorithm terminates when the optimal fitness no longer significantly improves over several generations. For example, if the optimal fitness value of the population stabilizes at around 0.85 after the 50th generation and shows no significant improvement for 10 consecutive generations, the algorithm can be considered converged.

[0077] Finally, the optimal solution is output. After the algorithm converges, the individual with the highest fitness value is selected from the population, and the airflow cell disruption parameter combination corresponding to this individual is the initial optimal airflow cell disruption parameter. For example, the final output initial optimal airflow cell disruption parameter combination is [1.3MPa, 23kg / h, 42℃, 3200r / min]. Verified by the pine pollen cell disruption quality predictor and the twin simulation model, the corresponding cell disruption quality coefficient is 93 points, the simulated cell disruption energy consumption is 120kW·h / ton, and the fitness value is 0.86. This effectively controls energy consumption while ensuring cell disruption quality, meeting the requirements of dual-objective optimization.

[0078] This step uses multiple iterations and population evolution of genetic learning algorithms to screen out the initial optimal airflow cell disruption parameters from a large number of parameter combinations that are suitable for material characteristics and balance quality and energy consumption, providing a scientific basis for the precise control of subsequent pine pollen airflow cell disruption processing.

[0079] S120: Control the airflow pulverizer to break the pine pollen to be processed according to the initial optimal airflow cell breaking parameters, and set the adaptation adjustment time window in the future period based on the actual cell breaking control parameter sequence obtained by real-time monitoring during the cell breaking process.

[0080] In this embodiment of the application, in order to dynamically adapt the airflow cell-wall breaking parameters to the changes in the state of pine pollen and the operation of the equipment during the processing, it is necessary to set a time window for subsequent parameter fine-tuning based on the actual cell-wall breaking control parameters monitored in real time, so as to improve the quality stability and energy economy of pine pollen airflow cell-wall breaking.

[0081] Specifically, the airflow pulverizer is first controlled to break down the pine pollen to be processed according to the obtained initial optimal airflow breaking down parameters. During the breaking down process, the actual breaking down control parameter sequence within a preset historical period is monitored and obtained in order to understand the actual fluctuation of the control parameters during the current breaking down process.

[0082] Furthermore, based on the initial optimal airflow wall-breaking parameters, the control deviation of the actual wall-breaking control parameter sequence is calculated to generate a control parameter deviation sequence, so as to show the degree of deviation between the actual parameters and the optimal parameters.

[0083] Meanwhile, the calculated mean control deviation is used as the overall control deviation, which can measure the average deviation level of the actual control parameters from the optimal parameters as a whole.

[0084] In addition, the ratio of the standard deviation of the parameters in the control parameter deviation sequence to the mean of the control deviation is set as the control fluctuation coefficient to reflect the severity of the control deviation fluctuation and to help judge the stability of the control process.

[0085] Finally, based on the current continuous processing time, overall control deviation, and control fluctuation coefficient, the time window compensation coefficient is calculated. After compensating the preset adjustment time window, the adaptive adjustment time window is obtained, thereby realizing the dynamic adaptation of the adjustment time window and providing a reasonable time rhythm basis for subsequent parameter fine-tuning.

[0086] Step S120 in the method provided in this application embodiment includes: During the cell wall breaking process, the actual cell wall breaking control parameter sequence within a preset historical time period is monitored and acquired, wherein the preset historical time period and the preset future time period have the same duration; Based on the initial optimal airflow cell disruption parameters, the control deviation of the actual cell disruption control parameter sequence is calculated to generate a control parameter deviation sequence, and the average control deviation is calculated as the overall control deviation. The ratio of the standard deviation of the parameters in the control parameter deviation sequence to the mean of the control deviation is set as the control fluctuation coefficient; Based on the current continuous processing time, the overall control deviation, and the control fluctuation coefficient, a time window compensation coefficient is calculated. After compensating the preset adjustment time window, an adaptive adjustment time window is obtained.

[0087] In this embodiment of the application, in order to avoid untimely or excessive adjustment of parameters during the pine pollen airflow cell breaking process due to a fixed time window, it is necessary to determine the appropriate adjustment time window for parameter fine-tuning based on factors such as the actual fluctuation of cell breaking control parameters and processing time during the processing process, so as to improve the stability and economy of pine pollen airflow cell breaking.

[0088] Specifically, the airflow pulverizer is first controlled to break down the pine pollen to be processed according to the initial optimal airflow breaking parameters, and the actual breaking control parameter sequence within a preset historical period is monitored and obtained during the pine pollen breaking process.

[0089] The actual cell-wall breaking control parameters also include key indicators such as airflow pressure, feed rate, crushing chamber temperature and classifying wheel speed. The length of the preset historical period must be the same as the preset future period to be set later, for example, both should be set to 10 minutes, so as to ensure that the adjustment window setting for the future period based on historical data is more referential.

[0090] For example, when the current cell-wall breaking process reaches the 20th minute, the actual cell-wall breaking control parameter sequence over the past 10 minutes (i.e., the 10th to 20th minutes) is monitored and obtained, where the airflow pressure fluctuates between 1.2 and 1.3 MPa, the feed rate fluctuates between 24 and 26 kg / h, etc.

[0091] Furthermore, based on the initial optimal airflow cell disruption parameters, the control deviation of the actual cell disruption control parameter sequence is calculated to generate a control parameter deviation sequence.

[0092] The initial optimal airflow cell disruption parameters were obtained through prior optimization, such as an airflow pressure of 1.25 MPa and a feed rate of 25 kg / h. When calculating the deviation between the actual cell disruption control parameters and the corresponding optimal airflow cell disruption parameters at each time point, if multiple parameters are involved at the same time point, the deviations of each parameter are weighted to obtain the comprehensive deviation value for that time point. The comprehensive deviation values ​​of all time points are then integrated to obtain the control parameter deviation sequence.

[0093] Simultaneously, the mean of the control parameter deviation sequence is calculated as the overall control deviation, which can measure the average deviation level of the actual cell disruption control parameters from the optimal airflow cell disruption parameters. For example, the overall control deviation calculated from the aforementioned 10-minute control parameter deviation sequence is 0.05 MPa.

[0094] Furthermore, the ratio of the standard deviation of the parameters to the mean of the control deviation in the control parameter deviation sequence is set as the control fluctuation coefficient to reflect the severity of the control deviation fluctuation, thereby helping to judge the stability of the control process. For example, if the standard deviation of the parameters is 0.03 MPa and the mean of the control deviation is 0.05 MPa, then the control fluctuation coefficient is 0.03 / 0.05 = 0.6, indicating that the deviation fluctuation is relatively stable.

[0095] Furthermore, based on the current continuous processing time, overall control deviation, and control fluctuation coefficient, a time window compensation coefficient is calculated. After compensating the preset adjustment time window, an adaptive adjustment time window is obtained.

[0096] The method provided in this application embodiment calculates a time window compensation coefficient based on the current continuous processing time, the overall control deviation, and the control fluctuation coefficient, including: The first window compensation coefficient is obtained based on the current continuous processing time, wherein the first window compensation coefficient gradually increases as the current continuous processing time increases; The ratio of the overall control deviation to the average historical overall control deviation within the historical time range is set as the second window compensation coefficient. The ratio of the control fluctuation coefficient to the average historical control fluctuation coefficient within the historical time range is set as the third window compensation coefficient. The time window compensation coefficient is obtained by weighted fusion of the first window compensation coefficient, the second window compensation coefficient, and the third window compensation coefficient.

[0097] Specifically, the first window compensation coefficient is obtained by matching the current continuous processing time.

[0098] Among them, the current continuous processing time is related to the state changes of pine pollen during the cell wall breaking process. That is, as the processing time increases, the cumulative amount of pine pollen broken and the equipment operating load may gradually change. It is necessary to adapt to these changes in a timely manner by increasing the fine-tuning frequency. Therefore, the first window compensation coefficient gradually increases with the increase of the current continuous processing time.

[0099] For example, the first window compensation coefficient is set to 0.6 for a continuous processing time of 0-1 hours; 0.8 for a continuous processing time of 1-2 hours; and 1.0 for a continuous processing time of more than 2 hours. By setting the gradient, a higher compensation coefficient can be obtained in the later stages of processing, providing support for shortening the adjustment time window and increasing the fine-tuning frequency.

[0100] Furthermore, the ratio of the overall control deviation to the average historical overall control deviation over the historical time range is set as the second window compensation coefficient.

[0101] Among them, the overall control deviation reflects the average deviation between the actual cell wall breaking parameters and the initial optimal parameters, while the historical average overall control deviation is the normal deviation level obtained by statistically analyzing multiple sets of pine pollen airflow cell wall breaking processing data. The ratio of the two can be used to determine whether the current deviation exceeds the normal range.

[0102] For example, if the historical average overall control deviation is 0.05 MPa and the current overall control deviation is 0.075 MPa, then the second window compensation coefficient is 0.075 / 0.05 = 1.5, indicating that the current deviation is higher than the normal level, and the time window needs to be adjusted by a higher compensation coefficient, and the parameter fine-tuning frequency needs to be increased to reduce the deviation; if the current overall control deviation is 0.04 MPa, then the second window compensation coefficient is 0.04 / 0.05 = 0.8, indicating that the current deviation is within the normal range, and the compensation coefficient can be appropriately reduced.

[0103] Meanwhile, the ratio of the control fluctuation coefficient to the average historical control fluctuation coefficient within the historical time range is set as the third window compensation coefficient.

[0104] Among them, the control fluctuation coefficient reflects the severity of parameter deviation fluctuation and the stability of the control process. The historical average control fluctuation coefficient is the normal level of deviation fluctuation in past processing. The ratio of the two can determine whether the current control stability meets the standard.

[0105] For example, if the historical average control fluctuation coefficient is 0.5 and the current control fluctuation coefficient is 0.75, the third window compensation coefficient is 0.75 / 0.5=1.2, indicating that the current parameter deviation fluctuation is relatively severe and the control stability is poor. It is necessary to increase the compensation coefficient to shorten the adjustment time window and enhance the timeliness of parameter fine-tuning. If the current control fluctuation coefficient is 0.4 and the third window compensation coefficient is 0.4 / 0.5=0.8, it indicates that the deviation fluctuation is mild and the control stability is good. The compensation coefficient can be appropriately lowered.

[0106] Furthermore, the time window compensation coefficient is obtained by weighted fusion of the first window compensation coefficient, the second window compensation coefficient, and the third window compensation coefficient.

[0107] Specifically, during weighted fusion, weights need to be set based on the core requirement of pine pollen airflow cell wall breaking. Considering the fundamental impact of processing time on the changes in pine pollen state, the first window compensation coefficient weight can be set to 0.4. Since the overall control deviation is directly related to the cell wall breaking quality, the second window compensation coefficient weight can be set to 0.3. Since control stability affects the accuracy of parameter adjustment, the third window compensation coefficient weight can be set to 0.3, and the sum of the three weights is 1.

[0108] For example, if the current continuous processing time is 2.5 hours (first window compensation coefficient 1.0), the overall control deviation is 0.075 MPa (second window compensation coefficient 1.5), and the control fluctuation coefficient is 0.75 (third window compensation coefficient 1.2), then the time window compensation coefficient = 1.0 × 0.4 + 1.5 × 0.3 + 1.2 × 0.3 = 1.21; if in another processing scenario, the continuous processing time is 0.5 hours (first window compensation coefficient 0.6), the overall control deviation is 0.04 MPa (second window compensation coefficient 0.8), and the control fluctuation coefficient is 0.4 (third window compensation coefficient 0.8), then the time window compensation coefficient = 0.6 × 0.4 + 0.8 × 0.3 + 0.8 × 0.3 = 0.72.

[0109] Furthermore, the calculated time window compensation coefficient is calculated and then used to adjust the preset time window. After compensating the preset time window, the adapted adjustment time window is obtained.

[0110] The preset adjustment time window is a fixed time window set based on the normal operating conditions of the air jet mill and the basic cell wall breaking requirements of pine pollen. For example, it is set to 2 minutes. This value needs to be determined with reference to the equipment characteristics of the air jet mill and the normal parameter adjustment interval for pine pollen cell wall breaking processing to ensure that the initial window meets the basic processing requirements.

[0111] Specifically, the compensation calculation formula can be expressed as "adaptive adjustment time window = preset adjustment time window / time window compensation coefficient". This formula dynamically corrects the window length through the compensation coefficient. When the time window compensation coefficient is greater than 1, it indicates that the current processing scenario requires more frequent parameter fine-tuning, and the window length is shortened by dividing by a coefficient greater than 1. When the compensation coefficient is less than 1, it indicates that the current processing state is more stable, and the window length is extended by dividing by a coefficient less than 1 to avoid excessive adjustment that causes parameter fluctuations.

[0112] For example, if the preset adjustment time window is 2 minutes, for the processing scenario where the above time window compensation coefficient is 1.21, the adapted adjustment time window = 2 / 1.21≈1.65 minutes, which means that parameters such as airflow pressure and feeding speed need to be fine-tuned every approximately 1.65 minutes. By shortening the window and increasing the adjustment frequency, deviations can be corrected in a timely manner to ensure the quality of cell wall breaking.

[0113] In addition, for processing scenarios with a time window compensation coefficient of 0.72, the appropriate adjustment time window is 2 / 0.72≈2.78 minutes. Extending the adjustment interval can avoid frequent parameter changes in the initial processing, reduce equipment energy waste and parameter instability risks, and at the same time conform to the actual processing characteristics of pine pollen with relatively stable material state in the initial stage of cell wall breaking.

[0114] Through the compensation methods described above, the resulting adaptation and adjustment time window can accurately match the real-time processing status of the pine pollen airflow cell wall breaking, laying a reasonable time rhythm foundation for subsequent parameter optimization and adjustment based on multi-source monitoring data.

[0115] S130: Within the adaptation and adjustment time window, the initial optimal airflow cell breaking parameters are optimized and adjusted according to the multi-source cell breaking monitoring data sequence composed of acoustic signals and machine vision data, and the current optimal airflow cell breaking parameters are output.

[0116] In this embodiment of the application, in order to achieve dynamic and precise optimization of the airflow cell-breaking parameters and avoid fluctuations in the cell-breaking effect caused by static parameter settings, it is necessary to optimize and adjust the initial optimal airflow cell-breaking parameters within the adaptation and adjustment time window, relying on the multi-source cell-breaking monitoring data sequence composed of acoustic signals and machine vision data, so as to ensure the high efficiency and stability of pine pollen cell-breaking.

[0117] Specifically, firstly, a sequence of adaptation adjustment time windows within a preset future time period is obtained, and the first adaptation adjustment time window in the sequence is selected as the first adaptation adjustment time window.

[0118] Furthermore, the system monitors and acquires the first acoustic signal sequence within the first adaptation adjustment time window, as well as the first visual monitoring data at the preset end node of the first adaptation adjustment time window.

[0119] The setting of the preset end node needs to be determined in conjunction with the time requirements of cell wall breaking parameter analysis to ensure the timeliness of parameter adjustment and avoid feedback adjustment delays.

[0120] After acquiring multi-source cell wall breaking monitoring data, cell wall breaking status identification and control error prediction are performed based on the first acoustic signal sequence and the first visual monitoring data. The initial optimal airflow cell wall breaking parameters are fine-tuned according to the predicted control error, and the current optimal airflow cell wall breaking parameters are output. The current optimal airflow cell wall breaking parameters are iteratively fine-tuned according to the adaptive adjustment time window sequence until the cell wall breaking process within the preset future time period is completed.

[0121] This step achieves dynamic optimization of airflow cell-wall breaking parameters through real-time feedback of multi-source cell-wall breaking monitoring data and iterative fine-tuning of parameters. It provides precise parameter control for the pine pollen airflow cell-wall breaking process, ensuring the consistency and efficiency of the cell-wall breaking effect.

[0122] Step S130 in the method provided in this application embodiment includes:

[0123] Obtain the sequence of adaptation adjustment time windows within a preset future time period, and select the first adaptation adjustment time window in the sequence as the first adaptation adjustment time window, and use the adjacent windows of the first adaptation adjustment time window as the second adaptation adjustment time window.

[0124] The system monitors and acquires the first acoustic signal sequence within the first adaptation adjustment time window, as well as the first visual monitoring data of the first adaptation adjustment time window at a preset end node.

[0125] Based on the first acoustic signal sequence and the first visual monitoring data, the cell wall breaking state is identified and the control error is predicted. The initial optimal airflow cell wall breaking parameters are fine-tuned according to the predicted control error, and the current optimal airflow cell wall breaking parameters are output. The current optimal airflow cell wall breaking parameters are iteratively fine-tuned according to the adaptive adjustment time window sequence until the cell wall breaking process within the preset future time period is completed.

[0126] In this embodiment of the application, in order to avoid the decline in cell wall breaking quality or the increase in energy consumption in the later stage of pine pollen airflow cell wall breaking due to fixed parameters, it is necessary to rely on the adaptation and adjustment time window sequence and dynamically optimize the parameters by combining acoustic and machine vision multi-source data to ensure the balance between pine pollen cell wall breaking quality and energy consumption.

[0127] Specifically, firstly, a sequence of adaptation adjustment time windows within a preset future time period is obtained, and the first adaptation adjustment time window in the sequence is selected as the first adaptation adjustment time window, and its adjacent windows are selected as the second adaptation adjustment time window.

[0128] The preset future time period needs to be determined by combining the total amount of pine pollen processed and the conventional processing efficiency of the air jet mill. For example, if the total processing volume needs to be completed in 2 hours, then the preset future time period is set to 2 hours. The adaptation adjustment time window sequence can be arranged in sequence according to the adaptation adjustment time windows calculated above to form multiple consecutive adjustment windows.

[0129] For example, if the first adaptation adjustment time window (first adaptation adjustment time window) is 9:00-9:01.65, the adjacent second adaptation adjustment time window is 9:01.65-9:03.3. The time window sequence ensures that parameter fine-tuning can cover the entire preset future time period and avoids adjustment gaps.

[0130] Furthermore, the system monitors and acquires the first acoustic signal sequence within the first adaptation adjustment time window, as well as the first visual monitoring data of the window at the preset end node.

[0131] The preset end node must be set in advance of the actual end node of the window to allow time for data processing and avoid feedback delays. For example, if the actual end node of the first adaptation adjustment time window is 9:01.65 and data processing takes 1 minute, then the preset end node is set to 9:00.65.

[0132] Specifically, when monitoring the first acoustic signal sequence, the sound signals generated by the breaking of pine pollen cells in the window are collected by an acoustic sensor. These signals contain acoustic information about particle collisions, friction, and impacts with the crushing chamber wall, which can reflect the intensity of the crushing.

[0133] Secondly, when monitoring the first visual monitoring data, the powder flow state image at the preset end node is captured by the online particle imaging device to obtain visual information such as powder flow concentration and particle morphology.

[0134] For example, within the first adaptation adjustment time window from 9:00 to 9:01.65, the acoustic sensor continuously collects an acoustic signal sequence with a frequency of 500-1000Hz and a sound pressure level of 80-85dB. At 9:00.65 (the preset end point), the online particle imaging device captures a visual image showing uniform powder flow concentration and particles with irregular and fine shapes. These data together constitute multi-source cell wall breakage monitoring data.

[0135] Furthermore, based on the first acoustic signal sequence and the first visual monitoring data, the cell wall breaking state is identified and the control error is predicted.

[0136] The method provided in this application embodiment, which performs cell wall breaking state identification and control error prediction based on the first acoustic signal sequence and the first visual monitoring data, includes: The first acoustic signal sequence is subjected to noise reduction filtering, time-frequency domain analysis and acoustic feature extraction in sequence to obtain the first key acoustic feature, wherein the key acoustic feature includes specific frequency band energy, main frequency / spectral centroid and sound pressure level; The first visual monitoring data is sequentially preprocessed and feature extracted to obtain the first key image features, wherein the key image features include powder flow concentration / density, particle size distribution and manifold morphology; Collect key acoustic feature sets, key image features, and control parameter error sets of samples; train a machine learning model until convergence; and build a control error prediction plugin. Using the control error prediction plugin, control error is predicted based on the first key acoustic feature and the first key image feature, and the predicted control error is output.

[0137] Specifically, the first acoustic signal sequence is first subjected to noise reduction filtering, time-frequency domain analysis and acoustic feature extraction in sequence to obtain the first key acoustic features.

[0138] The first acoustic signal sequence is the sound data of pine pollen cell wall breaking collected within the first adaptation adjustment time window. It includes acoustic information generated by particle collisions, friction and impacts with the crushing chamber wall. However, it will be mixed with environmental noise. It needs to be processed by noise reduction filtering to remove interference signals in order to retain the effective acoustic signals directly related to the cell wall breaking process.

[0139] Furthermore, time-frequency domain analysis is performed to convert the one-dimensional time-domain acoustic signal into a time-frequency diagram with a three-dimensional distribution of time, frequency, and energy. Methods such as short-time Fourier transform are then used to capture the frequency variation patterns of the acoustic signal over different time periods. For example, in the early stages of cell wall disruption, when the particles are larger, the proportion of low-frequency signals generated by collisions is high; as the particles become smaller, the proportion of high-frequency signals gradually increases.

[0140] Finally, specific frequency band energy, dominant frequency / spectral centroid, and sound pressure level are extracted from the processed acoustic signal as the first key acoustic features. Specifically, the specific frequency band energy can be selected from the 500-2000Hz band related to particle collisions, and the total energy within this band is calculated; higher energy generally indicates more intense pulverization.

[0141] Secondly, the dominant frequency is the frequency value where energy is most concentrated in the acoustic signal. For example, when the cell walls are fully broken down, the dominant frequency may stabilize at around 1200Hz. The sound pressure level reflects the intensity of the sound and can be directly collected and calibrated through a sound pressure sensor. For example, during normal cell wall breaking down, the sound pressure level is stable at 80-85dB.

[0142] Furthermore, the first visual monitoring data undergoes image preprocessing and feature extraction sequentially to obtain the first key image features. The first visual monitoring data consists of images of pine pollen powder flow captured by an online particle imaging device at the preset end node of the first adaptation adjustment time window, and requires prior image preprocessing.

[0143] Specifically, grayscale processing is used to convert the color image to a grayscale image to reduce color interference. Secondly, Gaussian filtering is used to remove image noise, making the edges of powder particles clearer. Finally, a threshold segmentation algorithm from existing technologies is used to separate the powder particles from the background, highlighting the effective target area.

[0144] Further, after preprocessing, feature extraction is performed to obtain powder flow concentration / density, particle size distribution, and manifold morphology as the first key image features.

[0145] The powder flow concentration / density is calculated by determining the proportion of pixels occupied by powder particles within a unit image area. Excessive concentration may indicate an excessively fast feed rate, potentially leading to insufficient cell disruption. Particle size distribution is determined by measuring the equivalent diameter of the particles using image analysis software and statistically analyzing the proportion of particles in different size ranges. If the proportion of 20-30μm particles exceeds 30%, it indicates that the current cell disruption effect has not met expectations.

[0146] In addition, the manifold morphology is determined by observing the outline characteristics of the particles. If the particles are irregular polyhedrals and there is no obvious agglomeration, it indicates that the cell wall breaking state is good. If the particles agglomerate into block-shaped manifold morphology, it may be due to the excessive temperature of the grinding chamber causing powder adhesion.

[0147] Furthermore, key acoustic feature sets, key image features, and control parameter error sets of the samples were collected to train a machine learning model until convergence. A control error prediction plugin was then constructed to achieve accurate mapping from multi-source monitoring data to control errors. The sample data were all derived from historical airflow cell-wall breaking processing records of pine pollen.

[0148] Specifically, the key acoustic feature set of the samples includes specific frequency band energy, dominant frequency, and sound pressure level data under different cell disruption states, such as specific frequency band energy of 1500J, dominant frequency of 1200Hz, and sound pressure level of 82dB. The key image feature set of the samples includes powder flow concentration, particle size distribution, and manifold morphology data under the corresponding states, such as powder flow concentration of 65%, 10-20μm particle ratio of 70%, and no agglomeration in manifold morphology. The sample control parameter error set consists of the deviation values ​​between the actual control parameters and the optimal airflow cell disruption parameters in the corresponding historical processing, such as airflow pressure deviation of -0.05MPa and feed rate deviation of +1kg / h.

[0149] Furthermore, a control error prediction plugin was built based on the random forest algorithm. This algorithm reduces the risk of overfitting by integrating the prediction results of multiple decision trees and can effectively handle the complex correlations between multi-source monitoring data.

[0150] Specifically, before model construction, the key acoustic feature set and key image feature set of the samples need to be preprocessed. For specific frequency band energy, dominant frequency, and sound pressure level in the key acoustic features of the samples, the Z-score standardization formula "standardized value = (original value - mean) / standard deviation" is used to eliminate magnitude differences; for powder flow concentration and the proportion of 10-20μm particles in the key image features of the samples, the percentage form is directly retained and converted to decimals.

[0151] Furthermore, unstructured features such as non-clustering manifold morphology need to be converted into numerical values ​​through label encoding. For example, "non-clustering" is denoted as 1, "slightly clustered" as 2, and "severely clustered" as 3, ensuring that all input features are in a numerical format recognizable by the model. Meanwhile, the sample control parameter error set does not require additional processing and can be directly used as a supervision signal for model training.

[0152] After preprocessing, the integrated sample feature set (i.e., standardized acoustic features + encoded image features) is used as input, and the sample control parameter error set is used as output. These are then divided into a training set and a validation set in a 7:3 ratio. The training set is used to construct the decision tree in the random forest, and the validation set is used to evaluate the model's generalization ability.

[0153] During model training, the number of decision trees is initially set to 100. Mean squared error is used as the evaluation metric for node splitting. Each decision tree is sampled from the training set based on Bootstrap sampling, and some features are randomly selected for node splitting to ensure the diversity of each decision tree.

[0154] After each training round, the average absolute error between the control error predicted by the model and the actual control error is calculated using the validation set. If the average absolute error of the validation set does not decrease for 10 consecutive rounds and remains stable within 0.02 MPa (airflow pressure deviation) and 0.5 kg / h (feed rate deviation), the model training is considered to have converged and the iteration is stopped.

[0155] Furthermore, the model performance needs to be verified using a test set (i.e., historical samples not used in training). If the average deviation between the predicted control error and the actual control error based on the key acoustic and image features of the samples in the test set is within a preset reasonable range, and the prediction results of multiple test samples are consistently in line with the accuracy requirements, then the model accuracy meets the requirements and can be determined as the final control error prediction plugin.

[0156] Furthermore, by using the trained control error prediction plugin, the first key acoustic feature and the first key image feature are input into it. After the internal algorithm of the control error prediction plugin is processed, the corresponding predicted control error is output to clarify the direction and magnitude of the deviation between the current wall breaking parameters and the optimal airflow wall breaking parameters.

[0157] Specifically, the first key acoustic feature and the first key image feature are preprocessed to ensure that the data format is consistent with the sample data during plugin training. Then, the two types of features are combined and input into the control error prediction plugin. The control error prediction plugin will calculate and output the predicted control error for airflow pressure, feed rate, crushing chamber temperature, and classifier wheel speed based on the feature-error correlation law learned during training.

[0158] Furthermore, the initial optimal airflow wall-breaking parameters are fine-tuned based on the obtained predictive control error to output the current optimal airflow wall-breaking parameters, and the current optimal airflow wall-breaking parameters are iteratively fine-tuned according to the adaptive adjustment time window sequence until the wall-breaking process within the preset future time period is completed.

[0159] Specifically, based on the predicted control error, reverse corrections are made for the airflow pressure, feed rate, crushing chamber temperature, and classifier speed. If the predicted airflow pressure deviation is negative, the actual airflow pressure is appropriately increased; if the feed rate deviation is positive, the actual feed rate is appropriately decreased. Through these targeted fine-tunings, the initial optimal airflow cell disruption parameters are corrected to the current optimal airflow cell disruption parameters that conform to the current processing state.

[0160] After the initial fine-tuning is completed, the next adaptation and adjustment time window is entered, and the process of "monitoring multi-source data - predicting control error - fine-tuning parameters" is repeated. That is, within this time window, new acoustic signal sequences and visual monitoring data are acquired again, and the control error prediction plugin is input to obtain new predictive control errors. Then, based on the new errors, the current optimal airflow wall breaking parameters are fine-tuned a second time.

[0161] Similarly, the process iterates sequentially according to the time window sequence, with each window dynamically adjusting parameters based on real-time monitoring data, until all pine pollen cell-wall breaking processes within the preset future time period are completed. This ensures that the current optimal airflow cell-wall breaking parameters are always precisely matched with the pine pollen state and equipment operation throughout the entire processing, maintaining stable cell-wall breaking quality and energy consumption levels.

[0162] For example, if the current optimal airflow pressure is 1.15 MPa and the feed rate is 24.1 kg / h after the first window is fine-tuned, after entering the second adaptation and adjustment time window, the newly monitored acoustic and visual features are input into the plug-in, predicting an airflow pressure deviation of +0.03 MPa and a feed rate deviation of -0.2 kg / h. Then, the airflow pressure is further fine-tuned to 1.12 MPa and the feed rate is fine-tuned to 24.3 kg / h. Through multiple iterations, the current optimal airflow cell disruption parameters are continuously optimized.

[0163] S140: Update the control command with the current optimal airflow cell-breaking parameters, continue to execute the cell-breaking process, and dynamically return to the cell-breaking process and monitoring steps based on the adaptive adjustment time window to perform closed-loop iterative regulation until all the pine pollen to be processed has been cell-breaking.

[0164] In this embodiment of the application, in order to ensure that the optimal airflow cell-breaking parameters after each parameter fine-tuning can be applied to the actual processing in a timely manner, it is necessary to convert the current optimal airflow cell-breaking parameters into continuous processing actions through instruction updates, equipment execution and dynamic feedback, and rely on the adaptation adjustment time window to achieve cyclic monitoring and control, so as to ensure that the pine pollen to be processed can complete cell-breaking under the adaptation parameters throughout the entire process.

[0165] Specifically, the first step is to establish the correlation between the current optimal airflow cell disruption parameters and the airflow pulverizer, and to clarify the equipment control object and operating standard corresponding to each parameter.

[0166] For example, an airflow pressure of 1.15 MPa corresponds to adjusting the opening of the compressed air valve of the airflow pulverizer, a feeding speed of 24.1 kg / h corresponds to adjusting the speed of the conveyor motor of the feeding device, a pulverizing chamber temperature of 42°C corresponds to controlling the temperature control module of the pulverizing chamber, and a classifying wheel speed of 3200 r / min corresponds to adjusting the frequency of the classifying wheel drive motor.

[0167] At the same time, the effective period of the current optimal airflow wall breaking parameters is marked according to the duration of the time window to ensure that the parameters work stably within the corresponding time window.

[0168] Furthermore, the current optimal airflow cell-wall breaking parameters are converted into control commands that the airflow pulverizer can recognize. The parameters are encoded according to the airflow pulverizer's communication protocol; for example, "airflow pressure 1.15MPa" is encoded as the hexadecimal command "0x01 0x03 0x00 0x01 0x00 0x01 0x84 0x0A". The encoded commands are then sent to the airflow pulverizer's control platform via industrial Ethernet to ensure that the equipment can start and operate according to the current optimal airflow cell-wall breaking parameters and continue processing pine pollen.

[0169] During the execution of the updated instructions by the equipment, dynamic monitoring and iterative control are carried out simultaneously based on the adaptation and adjustment time window. On the one hand, the operating parameters of the airflow pulverizer are collected in real time through the equipment's sensors, such as the airflow pressure sensor monitoring the actual airflow pressure and the speed sensor collecting the actual speed of the classifier wheel, to determine whether the equipment is operating normally according to the current optimal airflow cell breaking parameters. If a parameter deviates, a supplementary adjustment instruction is issued in a timely manner to correct it.

[0170] On the other hand, according to the adaptation and adjustment time window cycle, the process of "monitoring multi-source data - predicting control error - fine-tuning parameters" is re-entered. When the current adaptation and adjustment time window is about to end, the acoustic signal sequence and visual monitoring data within the time window are monitored and acquired again. The data are input into the control error prediction plugin to obtain a new predictive control error. Based on the new predictive control error, the current optimal airflow wall-breaking parameters are fine-tuned to generate new current optimal airflow wall-breaking parameters.

[0171] Furthermore, the control command is updated again with the new current optimal airflow cell disruption parameters and sent to the airflow pulverizer control platform. At the same time, based on the new adaptation adjustment time window (calculated from the new processing state), the above steps of "command update - equipment execution - monitoring - parameter fine-tuning" are repeated to form a closed-loop iterative control.

[0172] Simultaneously, it is necessary to continuously monitor the remaining amount of pine pollen to be processed and use a weighing sensor to determine whether further adjustments are needed. If the remaining amount of pine pollen is still greater than the minimum processing capacity of the equipment per cycle, then the closed-loop iteration continues according to the adapted adjustment time window; if the remaining amount of pine pollen is zero, or if all the cell wall breaking process has been completed, then adjustments are stopped.

[0173] Through the above steps, continuous closed-loop control of pine pollen airflow cell disruption from parameter optimization to actual processing is achieved, avoiding the limitations caused by processing with fixed parameters, ensuring that each processing stage can be carried out under suitable parameters, and guaranteeing the cell disruption quality and energy economy of the pine pollen throughout the entire process.

[0174] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects: This application proposes a multimodal data fusion method for optimizing pine pollen airflow cell disruption parameters. First, based on the material state data of the pine pollen to be processed and combined with the adjustment space of the airflow cell disruption parameters of the airflow mill, several pine pollen cell disruption schemes are randomly generated. Using a pre-constructed pine pollen cell disruption quality predictor and a twin simulation model of the airflow mill, the cell disruption quality coefficient and simulated energy consumption of each scheme are predicted. Then, the fitness of the scheme is calculated using a dual-objective parameter fitness evaluation function. Finally, an iterative optimization using a genetic learning algorithm is employed to output the initial optimal airflow cell disruption parameters. Next, the airflow mill is started according to the initial optimal airflow cell disruption parameters. During the process, the actual control parameter sequence within a preset historical period is monitored. Using the initial optimal airflow cell disruption parameters as a benchmark, the control deviation sequence, overall control deviation, and control fluctuation coefficient are calculated. Combined with the current continuous processing duration, a time window compensation coefficient is obtained. After compensating the preset adjustment time window, an adaptive adjustment time window is generated. Subsequently, within the adaptation and adjustment time window, acoustic signal sequences and machine vision data are collected and input into the control error prediction plugin to obtain the predicted control error. The initial optimal airflow cell-breaking parameters are then fine-tuned to obtain the current optimal airflow cell-breaking parameters, and this fine-tuning is iteratively performed according to the adaptation and adjustment time window sequence until the processing for the preset future time period is completed. Finally, the control command is updated with the current optimal airflow cell-breaking parameters, and the cell-breaking process is continuously executed. Relying on the adaptation and adjustment time window, the closed-loop iteration of the "command update - equipment execution - monitoring - parameter fine-tuning" steps is dynamically returned, and the equipment operation deviation is corrected in real time until all the pine pollen to be processed is cell-broken.

[0175] The method provided in this application solves the problems of traditional pine pollen airflow cell breaking relying on fixed parameters, poor adaptability, and difficulty in balancing cell breaking quality and energy consumption by adopting the technical solution of "dual-objective parameter optimization - dynamic window setting - real-time control of multi-source data - full-process closed-loop iteration". It achieves bidirectional optimization of quality and energy consumption in the entire process of pine pollen airflow cell breaking, and provides efficient technical support for the high-value processing of pine pollen.

[0176] Example 2, as shown in the appendix Figure 2 As shown, based on the inventive concept of the pine pollen airflow cell disruption parameter optimization method using multimodal data fusion provided in Embodiment 1, this application also provides a pine pollen airflow cell disruption parameter optimization system using multimodal data fusion, specifically including: The initial parameter optimization module 01 is used to optimize the initial optimal airflow cell breaking parameters based on the material state data of the pine pollen to be processed, within the airflow cell breaking parameter adjustment space, with the dual objectives of maximizing cell breaking quality and minimizing equipment energy consumption.

[0177] The adaptation window setting module 02 is used to control the airflow pulverizer to break the pine pollen to be processed according to the initial optimal airflow cell breaking parameters, and to set an adaptation adjustment time window in the future period based on the actual cell breaking control parameter sequence obtained by real-time monitoring during the cell breaking process.

[0178] The multi-source data adjustment module 03 is used to optimize and adjust the initial optimal airflow cell breaking parameters according to the multi-source cell breaking monitoring data sequence composed of acoustic signals and machine vision data within the adaptation adjustment time window, and output the current optimal airflow cell breaking parameters.

[0179] The closed-loop iterative control module 04 is used to update the control command with the current optimal airflow cell-breaking parameters, continue to execute the cell-breaking process, and dynamically return to the cell-breaking process and monitoring steps based on the adaptive adjustment time window to perform closed-loop iterative control until all the pine pollen to be processed has been cell-breaking.

[0180] In one embodiment, the initial parameter optimization module 01 is further configured to: The initial particle size distribution, toughness, hardness, and moisture content of the pine pollen to be processed are collected and obtained as material state data. The airflow cell disruption parameter adjustment space of the airflow mill is obtained, wherein the airflow cell disruption parameters include airflow pressure, feed rate, milling chamber temperature, and classifier wheel speed. Several initial airflow cell disruption parameters are randomly selected within the airflow cell disruption parameter adjustment space and combined with the material state data to generate several pine pollen cell disruption schemes. Using a pre-constructed pine pollen cell disruption quality predictor and a twin simulation model of the airflow mill, several cell disruption quality coefficients and several simulated cell disruption energy consumptions are predicted based on the several pine pollen cell disruption schemes. A parameter fitness evaluation function is constructed with the dual objectives of maximizing cell disruption quality and minimizing equipment energy consumption. Several parameter fitnesss are calculated based on the several cell disruption quality coefficients and several simulated cell disruption energy consumptions. A genetic learning algorithm is used to optimize based on the several initial airflow cell disruption parameters and several parameter fitnesss, outputting the initial optimal airflow cell disruption parameters.

[0181] Furthermore, the initial parameter optimization module 01 also includes: Based on historical airflow cell wall disruption processing records of pine pollen, a sample cell wall disruption scheme set and a sample cell wall disruption quality dataset were collected. The sample cell wall disruption quality data were evaluated to determine the sample cell wall disruption quality coefficient, and a sample cell wall disruption quality coefficient set was obtained. The cell wall disruption quality data includes at least the cell wall disruption rate, particle size distribution, and key nutrient retention rate. Using the sample cell wall disruption scheme set as input, and the sample cell wall disruption quality dataset as supervised training, a deep learning model was trained until convergence to generate a pine pollen cell wall disruption quality predictor.

[0182] In one embodiment, the adapter window setting module 02 is further configured to: During the cell wall breaking process, the actual cell wall breaking control parameter sequence within a preset historical period is monitored and acquired, wherein the preset historical period and the preset future period have the same duration. Based on the initial optimal airflow cell wall breaking parameters, the control deviation of the actual cell wall breaking control parameter sequence is calculated to generate a control parameter deviation sequence, and the mean value of the control deviation is calculated as the overall control deviation. The ratio of the standard deviation of the parameters in the control parameter deviation sequence to the mean value of the control deviation is set as the control fluctuation coefficient. Based on the current continuous processing time, the overall control deviation, and the control fluctuation coefficient, a time window compensation coefficient is calculated, and after compensating the preset adjustment time window, an adaptive adjustment time window is obtained.

[0183] Furthermore, the adaptation window setting module 02 also includes: The first window compensation coefficient is obtained based on the current continuous processing duration, wherein the first window compensation coefficient gradually increases with the increase of the current continuous processing duration; the ratio of the overall control deviation to the average historical overall control deviation within the historical time range is set as the second window compensation coefficient; the ratio of the control fluctuation coefficient to the average historical control fluctuation coefficient within the historical time range is set as the third window compensation coefficient; the time window compensation coefficient is obtained by weighted fusion of the first window compensation coefficient, the second window compensation coefficient, and the third window compensation coefficient.

[0184] In one embodiment, the multi-source data adjustment module 03 is further configured to: Obtain a sequence of adaptation adjustment time windows within a preset future time period, and select the first adaptation adjustment time window in the sequence as the first adaptation adjustment time window, and use the adjacent windows of the first adaptation adjustment time window as the second adaptation adjustment time window; monitor and obtain the first acoustic signal sequence within the first adaptation adjustment time window, and the first visual monitoring data of the first adaptation adjustment time window at a preset end node; based on the first acoustic signal sequence and the first visual monitoring data, identify the wall breaking state and predict the control error, fine-tune the initial optimal airflow wall breaking parameters according to the predicted control error, output the current optimal airflow wall breaking parameters, and perform iterative fine-tuning of the current optimal airflow wall breaking parameters according to the adaptation adjustment time window sequence until the wall breaking process within the preset future time period is completed.

[0185] Furthermore, the multi-source data adjustment module 03 also includes: The first acoustic signal sequence is sequentially subjected to noise reduction filtering, time-frequency domain analysis, and acoustic feature extraction to obtain the first key acoustic feature, wherein the key acoustic feature includes specific frequency band energy, dominant frequency / spectral centroid, and sound pressure level; the first visual monitoring data is sequentially subjected to image preprocessing and feature extraction to obtain the first key image feature, wherein the key image feature includes powder flow concentration / density, particle size distribution, and manifold morphology; a machine learning model is trained to convergence by collecting the sample key acoustic feature set, sample key image feature, and sample control parameter error set to construct a control error prediction plugin; using the control error prediction plugin, control error is predicted based on the first key acoustic feature and the first key image feature, and the predicted control error is output.

[0186] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0187] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0188] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for optimizing parameters of pollen airflow de-budding by multi-modal data fusion, characterized in that the method comprises the following steps: The method comprises the following steps: ​ Based on the material state data of the pine pollen to be processed, the initial optimal airflow breaking wall parameters are determined in the airflow breaking wall parameter adjustment space, with the dual objectives of maximizing the breaking wall quality and minimizing the equipment energy consumption. The initial optimal airflow breaking wall parameters are used to control the airflow crusher to break the wall of the pine pollen to be processed, and during the breaking wall process, an adaptive adjustment time window in a preset future period is set based on the actual breaking wall control parameter sequence obtained through real-time monitoring. In the adaptive adjustment time window, the initial optimal airflow breaking wall parameters are optimized and adjusted according to the multi-source breaking wall monitoring data sequence composed of acoustic signals and machine vision data, and the current optimal airflow breaking wall parameters are output. The control instruction is updated with the current optimal airflow breaking wall parameters, and the breaking wall process is continued. The breaking wall processing and monitoring steps are dynamically returned for closed-loop iterative control based on the adaptive adjustment time window, until the pine pollen to be processed is completely broken.

2. The multi-modal data fusion's pine pollen air flow broken wall parameter optimization method according to claim 1, characterized in that, Based on the material state data of the pine pollen to be processed, the initial optimal airflow breaking wall parameters are determined in the airflow breaking wall parameter adjustment space, with the dual objectives of maximizing the breaking wall quality and minimizing the equipment energy consumption, comprising: The initial particle size distribution, toughness, hardness and water content of the pine pollen to be processed are collected and obtained as material state data; The airflow breaking wall parameter adjustment space of the airflow crusher is obtained, wherein the airflow breaking wall parameters include airflow pressure, feeding speed, crushing chamber temperature and grading wheel rotating speed; A plurality of initial airflow breaking wall parameters are randomly selected in the airflow breaking wall parameter adjustment space, which are combined with the material state data to generate a plurality of pine pollen breaking wall schemes; A pine pollen breaking wall quality predictor and a twin simulation model of the airflow crusher are constructed in advance, and a plurality of breaking wall quality coefficients and a plurality of breaking wall model energy consumptions are predicted according to the plurality of pine pollen breaking wall schemes; A parameter fitness evaluation function is constructed with the dual objectives of maximizing the breaking wall quality and minimizing the equipment energy consumption, and a plurality of parameter fitnesses are calculated according to the plurality of breaking wall quality coefficients and the plurality of breaking wall model energy consumptions; A genetic learning algorithm is used to optimize the plurality of initial airflow breaking wall parameters and the plurality of parameter fitnesses, and the initial optimal airflow breaking wall parameters are output.

3. The multi-modal data fusion's pine pollen air flow broken wall parameter optimization method according to claim 2, characterized in that, The construction steps of the pine pollen breaking wall quality predictor comprise: Based on the historical airflow breaking wall processing records of pine pollen, a sample breaking wall scheme set and a sample breaking wall quality data set are collected, and the sample breaking wall quality data is evaluated to determine the sample breaking wall quality coefficients, and a sample breaking wall quality coefficient set is obtained, wherein the breaking wall quality data at least includes the breaking rate, the particle size distribution and the key nutrient retention rate; The sample breaking wall scheme set is used as the input, and the sample breaking wall quality data set is used as the supervision to train the deep learning model to convergence, and the pine pollen breaking wall quality predictor is generated.

4. The multi-modal data fusion's pine pollen air flow broken wall parameter optimization method according to claim 1, characterized in that, The adaptive adjustment time window in a preset future period is set based on the actual breaking wall control parameter sequence obtained through real-time monitoring, comprising: In the process of breaking the wall, the actual breaking wall control parameter sequence in the preset historical period is monitored and obtained, wherein the preset historical period and the preset future period have the same time length; Based on the initial optimal airflow breaking wall parameter, control deviation calculation is performed on the actual breaking wall control parameter sequence to generate a control parameter deviation sequence, and a control deviation mean value is calculated as the overall control deviation; The ratio of the parameter standard deviation in the control parameter deviation sequence to the control deviation mean value is set as a control fluctuation coefficient; Based on the current continuous processing time length, the overall control deviation and the control fluctuation coefficient, a time window compensation coefficient is calculated, and after compensating for the preset adjustment time window, an adaptive adjustment time window is obtained.

5. The multi-modal data fusion's pine pollen air flow broken wall parameter optimization method according to claim 4, characterized in that, Based on the current continuous processing time length, the overall control deviation and the control fluctuation coefficient, a time window compensation coefficient is calculated, and after compensating for the preset adjustment time window, an adaptive adjustment time window is obtained. Based on the current continuous processing time length, the overall control deviation and the control fluctuation coefficient, a time window compensation coefficient is calculated, and after compensating for the preset adjustment time window, an adaptive adjustment time window is obtained. Based on the current continuous processing time length, the overall control deviation and the control fluctuation coefficient, a time window compensation coefficient is calculated, and after compensating for the preset adjustment time window, an adaptive adjustment time window is obtained. According to the first window compensation coefficient, the second window compensation coefficient and the third window compensation coefficient, a time window compensation coefficient is obtained. In the adaptive adjustment time window, the initial optimal airflow breaking wall parameter is optimized and adjusted according to the multi-source breaking wall monitoring data sequence composed of acoustic signals and machine vision data, and the current optimal airflow breaking wall parameter is output, including:

6. The multi-modal data fusion's pine pollen air flow broken wall parameter optimization method of claim 1, wherein, An adaptive adjustment time window sequence in a preset future period is obtained, and the first adaptive adjustment time window in the adaptive adjustment time window sequence is selected as a first adaptive adjustment time window, and the adjacent window of the first adaptive adjustment time window is selected as a second adaptive adjustment time window; A first acoustic signal sequence in the first adaptive adjustment time window is monitored and obtained, and a first visual monitoring data under a preset end node of the first adaptive adjustment time window is monitored and obtained; Based on the first acoustic signal sequence and the first visual monitoring data, breaking wall state recognition and control error prediction are performed, the initial optimal airflow breaking wall parameter is fine-tuned according to the predicted control error, the current optimal airflow breaking wall parameter is output, and the iterative fine-tuning of the current optimal airflow breaking wall parameter is performed according to the adaptive adjustment time window sequence until the breaking wall processing in the preset future period is completed. Based on the first acoustic signal sequence and the first visual monitoring data, breaking wall state recognition and control error prediction are performed, including:

7. The multi-modal data fusion's pine pollen air flow broken wall parameter optimization method according to claim 6, characterized in that, The first acoustic signal sequence is sequentially subjected to noise reduction filtering, time-frequency domain analysis and acoustic feature extraction to obtain a first key acoustic feature, wherein the key acoustic feature includes a specific frequency band energy, a main frequency / spectral barycenter and a sound pressure level; ​ The first visual monitoring data is sequentially subjected to image preprocessing and feature extraction to obtain first key image features, wherein the key image features include powder flow concentration / density, particle size distribution, and flow morphology; A machine learning model is trained to convergence using a sample key acoustic feature set, a sample key image feature, and a sample control parameter error set, and a control error prediction plug-in is constructed; The control error prediction plug-in is used to predict control errors based on the first key acoustic features and the first key image features, and to output predicted control errors.

8. A multi-modal data fusion's pine pollen airflow broken wall parameter optimization system, characterized in that, The system is used to perform the multi-modal data fusion pine pollen airflow wall-breaking parameter optimization method of any one of claims 1-7, and the system comprises: An initial parameter optimization module is configured to perform optimization based on material state data of pine pollen to be processed in an airflow wall-breaking parameter adjustment space, with the dual objectives of maximizing wall-breaking quality and minimizing equipment energy consumption, to determine initial optimal airflow wall-breaking parameters; An adaptive window setting module is configured to control an airflow crusher to perform wall-breaking processing on the pine pollen to be processed according to the initial optimal airflow wall-breaking parameters, and to set an adaptive adjustment time window in a preset future period based on an actual wall-breaking control parameter sequence obtained through real-time monitoring during the wall-breaking processing; A multi-source data adjustment module is configured to optimize and adjust the initial optimal airflow wall-breaking parameters according to a multi-source wall-breaking monitoring data sequence composed of acoustic signals and machine vision data within the adaptive adjustment time window, and to output current optimal airflow wall-breaking parameters; A closed-loop iterative control module is configured to update control instructions with the current optimal airflow wall-breaking parameters, to continue performing wall-breaking processing, and to perform closed-loop iterative control based on the adaptive adjustment time window by dynamically returning to the wall-breaking processing and monitoring steps until the pine pollen to be processed is completely wall-broken.

Images