Dry fermentation fermentation data processing method, device and computer equipment

Abstract

The embodiment of the present specification relates to the technical field of fermentation, in particular to a dry fermentation data processing method and device and computer equipment. The method comprises: training a first fermentation product prediction model constructed based on at least two model construction methods, the first fermentation product prediction model being used for predicting fermentation products according to material parameters and operation parameters in a dry fermentation process; determining performance indicators of the trained at least two first fermentation product prediction models; selecting a target fermentation product prediction model from the at least two first fermentation product prediction models according to the performance indicators; and predicting fermentation products according to the target fermentation product prediction model. The embodiment of the present specification can predict fermentation products in the dry fermentation process, which is of great significance to guide the safe operation and efficient production of the factory.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of fermentation technology, and in particular to a method, apparatus, and computer equipment for processing fermentation data during dry fermentation. Background Technology

[0002] Dry fermentation is an important method for the harmless treatment of organic solid waste and the simultaneous production of clean energy biogas. Compared with traditional wet anaerobic fermentation, dry fermentation has advantages such as smaller reactor volume, lower energy input, water conservation, and easier treatment of biogas slurry and residue. In water-scarce areas, it is extremely beneficial for the harmless treatment of organic waste resources. Fermentation products are the main production indicators for industrial-scale garage-type dry fermentation plants. Accurate prediction is crucial for guiding the safe operation and efficient production of the plant. Summary of the Invention

[0003] This specification provides an embodiment of a fermentation data processing method, apparatus, and computer device for dry fermentation.

[0004] This specification provides an embodiment of a method for processing fermentation data during dry fermentation, including:

[0005] A first fermentation product prediction model, constructed based on at least two model construction methods, is trained. The first fermentation product prediction model is used to predict fermentation products based on material parameters and operating parameters during dry fermentation.

[0006] Determine the performance metrics of at least two first fermentation product prediction models after training.

[0007] Select the target fermentation product prediction model from at least two first fermentation product prediction models based on performance indicators;

[0008] The fermentation products are predicted based on the target fermentation product prediction model.

[0009] This specification also provides an embodiment of a fermentation data processing device for dry fermentation, comprising:

[0010] The training unit is used to train a first fermentation product prediction model constructed based on at least two model construction methods. The first fermentation product prediction model is used to predict fermentation products based on material parameters and operating parameters of dry fermentation.

[0011] A determination unit is used to determine the performance metrics of at least two first fermentation product prediction models after training.

[0012] The selection unit is used to select a target fermentation product prediction model from at least two first fermentation product prediction models based on performance indicators;

[0013] The prediction unit is used to predict the fermentation products based on the target fermentation product prediction model.

[0014] This specification also provides an embodiment of a computer device, including:

[0015] processor;

[0016] Memory used to store processor-executable instructions;

[0017] The processor executes the instructions to implement a fermentation data processing method for dry fermentation.

[0018] The technical solution provided in the embodiments of this specification can train a first fermentation product prediction model constructed based on at least two model construction methods. This first fermentation product prediction model is used to predict fermentation products based on material parameters and operating parameters during dry fermentation. The performance indicators of the at least two trained first fermentation product prediction models can be determined. A target fermentation product prediction model can be selected from the at least two first fermentation product prediction models based on the performance indicators. Fermentation products can be predicted based on the target fermentation product prediction model. This allows for the prediction of fermentation products during dry fermentation, which is of great significance for guiding the safe operation and efficient production of a plant. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. The drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the fermentation data processing method for dry fermentation in the embodiments of this specification.

[0021] Figure 2 This is a schematic diagram of the fermentation data processing process for dry fermentation in the embodiments of this specification;

[0022] Figure 3 This is a schematic diagram illustrating the process of determining the values ​​of operating parameters using the particle swarm optimization algorithm in the embodiments of this specification.

[0023] Figure 4A This diagram illustrates the changes in prediction performance of the AutoGluon-based model in the embodiments of this specification over a runtime range of 60-1200 seconds.

[0024] Figure 4BThis is a schematic diagram illustrating the changes in prediction performance of the H2O-based model in the embodiments of this specification within a runtime range of 60-1200s;

[0025] Figure 4C This is a schematic diagram illustrating the predictive performance of the AutoGluon method-based model for biogas production in a digester in the embodiments of this specification.

[0026] Figure 4D This is a schematic diagram illustrating the predictive performance of the AutoGluon method-based model for methane content in a digester in the embodiments of this specification.

[0027] Figure 5A This is a schematic diagram showing the SHAP value analysis results of biogas production from the digester in the embodiments of this specification.

[0028] Figure 5B This is a schematic diagram showing the SHAP value analysis results of methane content in the digester in the embodiments of this specification;

[0029] Figure 5C is a schematic diagram of the SHAP value analysis results of digester temperature and biogas production in the embodiments of this specification;

[0030] Figure 5D This is a schematic diagram showing the SHAP value analysis results of spraying time and biogas production in the embodiments of this specification.

[0031] Figure 5E This is a schematic diagram showing the SHAP value analysis results of digester pressure and biogas production in the embodiments of this specification;

[0032] Figure 5F This is a schematic diagram showing the SHAP value analysis results of raw material quality and methane content in the examples of this specification;

[0033] Figure 5G This is a schematic diagram showing the SHAP value analysis results of digester temperature and methane content in the embodiments of this specification;

[0034] Figure 5H This is a schematic diagram showing the SHAP value analysis results of the total solids content and methane content of the raw materials in the examples of this specification;

[0035] Figure 6A This diagram illustrates the changes in prediction performance of the AutoGluon-based model in the embodiments of this specification over a runtime range of 60-1200 seconds.

[0036] Figure 6B This is a schematic diagram illustrating the changes in prediction performance of the H2O-based model in the embodiments of this specification within a runtime range of 60-1200s;

[0037] Figure 6CThis is a schematic diagram illustrating the predictive performance of the AutoGluon method-based model for biogas production in leachate ponds in the embodiments of this specification.

[0038] Figure 6D This is a schematic diagram illustrating the predictive performance of the AutoGluon method-based model for methane content in leachate ponds in the embodiments of this specification.

[0039] Figure 7A This is a schematic diagram showing the SHAP value analysis results of the biogas production of the leachate tank in the embodiments of this specification.

[0040] Figure 7B This is a schematic diagram showing the SHAP value analysis results of the methane content in the leachate tank in the embodiments of this specification.

[0041] Figure 7C This is a schematic diagram showing the SHAP value analysis results of leachate tank temperature and biogas production in the embodiments of this specification;

[0042] Figure 7D This is a schematic diagram showing the SHAP value analysis results of liquid level and biogas production in the embodiments of this specification;

[0043] Figure 7E This is a schematic diagram showing the analysis results of pH and biogas production SHAP values ​​in the examples of this specification;

[0044] Figure 7F This is a schematic diagram showing the SHAP value analysis results of the leachate tank temperature and methane content in the embodiments of this specification;

[0045] Figure 7G This is a schematic diagram showing the SHAP value analysis results of liquid level and methane content in the embodiments of this specification;

[0046] Figure 7H This is a schematic diagram showing the SHAP value analysis results of total inorganic carbon and methane content in the examples of this specification;

[0047] Figure 8A This is a schematic diagram showing the test results of biogas production in the digester in the embodiments of this specification;

[0048] Figure 8B This is a schematic diagram showing the test results of methane content in the digester in the embodiments of this specification;

[0049] Figure 8C This is a schematic diagram showing the test results of biogas production from the leachate tank in the embodiments of this specification;

[0050] Figure 8D This is a schematic diagram showing the test results of methane content in the leachate tank in the embodiments of this specification;

[0051] Figure 9This is a schematic diagram of the fermentation data processing device for dry fermentation in the embodiments of this specification. Detailed Implementation

[0052] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. The specific embodiments described herein are only used to explain this disclosure, and not to limit this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure are within the scope of protection of this disclosure. In addition, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0053] Currently, the application scope of industrial-scale dry fermentation technology is gradually expanding. However, the prediction and early warning guidance of fermentation products have not yet been applied to actual production. The reasons include the following aspects: (1) Dry fermentation needs to overcome some disadvantages caused by high total solids content; (2) There are many factors affecting dry fermentation, including hydrogen partial pressure, temperature, pH value, oxygen content, redox potential, water content, water activity, etc.; (3) There are many types of dry fermentation materials and operating parameters, and the correlation between parameters is unclear. It is difficult to grasp the influence of material property parameters and operating parameters on the biogas production performance of dry fermentation; (4) Dry fermentation has highly complex, nonlinear and dynamic characteristics, and is very easily affected by uncertain factors. It is difficult to simulate and study using traditional kinetic methods. The above aspects together limit the large-scale utilization of industrial-scale garage-type dry fermentation as a high-value treatment method for organic solid waste. In addition, the internal environment of the dry fermentation reactor is complex and variable. The operating parameters of dry fermentation are highly nonlinear and unbalanced, and it is also difficult to effectively control the operating parameters using traditional machine learning methods to maintain the stable reaction. This will cause some problems, such as low prediction accuracy, overfitting, acidification of reactants, and low gas production performance.

[0054] This specification provides a method for processing fermentation data during dry fermentation. The method can be applied to computer equipment such as servers. This specification does not limit the type or structure of the computer equipment, as long as it can implement the method steps described in this specification. For example, multiple method steps can be implemented by a single server, or the multiple method steps can be distributed across multiple servers.

[0055] Please see Figure 1 and Figure 2 The method may include the following steps.

[0056] Step 11: Train the first fermentation product prediction model constructed based on at least two model construction methods. The first fermentation product prediction model is used to predict the fermentation products based on the material parameters and operating parameters in the dry fermentation process.

[0057] In some embodiments, the at least two model building methods may include the AutoGluon method and the H2O method, etc. The AutoGluon method and the H2O method may include automated machine learning methods (AutoML).

[0058] In some embodiments, at least two first fermentation product prediction models can be constructed based on the at least two model construction methods. Each model construction method can construct one first fermentation product prediction model.

[0059] The first fermentation product prediction model is used to predict fermentation products based on material parameters and operational parameters during dry fermentation. The first fermentation product prediction model may include an ensemble model. The ensemble model is obtained by integrating multiple sub-models. These sub-models have strong learning capabilities and significant differences in structural performance. The integration method of the multiple sub-models may include multi-level stacking strategies, etc. For example, the ensemble model is obtained by averaging, weighted averaging, voting, stacking, etc., of the outputs of the multiple sub-models. The sub-models may include at least one of the following: LightGBM (Lightweight Gradient Boosting Machine), CatBoost (Gradient Boosting and Categorical Features), ExtraTrees (Limited Tree), Random Forest (RF), NeuralNet, Extreme Gradient Boosting Tree (XGBoost), K-Nearest Neighbors (KNN), etc. The ensemble model may include a WeightedEnsemble model, etc.

[0060] The first fermentation product prediction model may include a multi-objective fermentation product prediction model. The output of the multi-objective fermentation product prediction model may include multiple predicted values ​​for fermentation products, thereby enabling the prediction of multiple fermentation products. For example, the output of the multi-objective fermentation product prediction model may include biogas production and methane content.

[0061] The AutoGluon method can automatically identify multi-class classification task datasets, perform automatic feature engineering, train multiple sub-models, and then integrate the better-performing sub-models to obtain a first fermentation product prediction model (e.g., WeightedEnsemble).

[0062] The H2O method combines fast random search and stacked ensembles to train various models to obtain a first fermentation product prediction model. Examples include random forests, extreme gradient boosting trees, gradient boosting machines, generalized linear models, and deep neural networks.

[0063] In some embodiments, the first fermentation product prediction models constructed by the at least two model construction methods can be trained based on a first sample dataset. The first sample dataset corresponds to a digester. This allows for the acquisition of at least two first fermentation product prediction models applicable to the digester through training.

[0064] The digester data may include raw material quality, temperature, pressure, spraying time, spraying frequency, total solids content of raw materials, volatile solids content of raw materials, total solids content of biogas residue, volatile solids content of biogas residue, biogas residue quality, and fermentation product values. Therefore, the first sample dataset may include at least one first sample data set. The first sample data set may include material parameter values, operating parameter values, and fermentation product values. The material parameters may include total solids content of raw materials, volatile solids content of raw materials, total solids content of biogas residue, volatile solids content of biogas residue, and biogas residue quality. The operating parameters may include raw material quality, temperature, pressure, spraying time, and spraying frequency. The fermentation products may include biogas and methane. The fermentation product values ​​may include biogas yield and methane content, which can be used as labels for the first sample data set.

[0065] In some scenario examples, the first sample data in the first sample dataset can be as shown in Table 1 below.

[0066] Table 1

[0067]

[0068] In Table 1 above, input variables may include raw material quality, digester temperature, digester pressure, spraying time, number of sprays, total solids content of raw materials, volatile solids content of raw materials, total solids content of biogas residue, volatile solids content of biogas residue, and biogas residue quality. Output variables may include biogas production and methane content.

[0069] In practical applications, for each first fermentation product prediction model, the first sample data from the first sample dataset can be input into the model to obtain its output. Based on the output and the labels of the first sample data, loss information can be calculated using a loss function. The parameters of the first fermentation product prediction model can then be adjusted based on the loss information, thus enabling the training of the model. The loss function includes logarithmic loss and square loss, among others.

[0070] In some embodiments, the first fermentation product prediction models constructed by the at least two model construction methods can be trained based on a second sample dataset. The second sample dataset corresponds to the leachate tank. This allows for the acquisition of at least two first fermentation product prediction models applicable to the leachate tank through training.

[0071] The leachate tank data may include temperature, pressure, liquid level, pH value, conductivity, total inorganic carbon content, volatile fatty acids, fermentation product values, etc. Therefore, the second sample dataset may include at least one second sample data set. The second sample data set may include material parameter values, operating parameter values, and fermentation product values. The material parameters may include conductivity, total inorganic carbon content, volatile fatty acids, etc. The operating parameters may include temperature, pressure, liquid level, pH value, etc. The fermentation products may include biogas and methane. The fermentation product values ​​may include biogas yield and methane content. The fermentation product values ​​can be used as labels for the second sample data set.

[0072] In some scenario examples, the second sample data in the second sample dataset can be as shown in Table 2 below.

[0073] Table 2

[0074]

[0075] In Table 2 above, input variables may include leachate tank temperature, leachate tank pressure, liquid level, pH, conductivity, total inorganic carbon, and volatile fatty acids. Output variables may include biogas production and methane content.

[0076] In practical applications, for each first fermentation product prediction model, the second sample data from the second sample dataset can be input into the first fermentation product prediction model to obtain its output. Loss information can be calculated using a loss function based on the output of the first fermentation product prediction model and the labels of the second sample data. The parameters of the first fermentation product prediction model can then be adjusted based on the loss information, thus achieving the training of the first fermentation product prediction model. The loss function includes logarithmic loss function, square loss function, etc.

[0077] Step 12: Determine the performance metrics of at least two first fermentation product prediction models after training.

[0078] In some embodiments, the performance metrics are used to represent the performance of the first fermentation product prediction model. The performance metrics include at least one of the following: accuracy, recall, F1 score, measured coefficients, mean absolute error, etc. The measured coefficients can be calculated according to the formula... The mean absolute error is calculated using the formula. Calculated. p i and o i These represent the actual value and the predicted value, respectively. The mean value represents the actual values. n represents the number of predicted values ​​(or actual values). The actual values ​​include labels, and the predicted values ​​include the output of the first fermentation product prediction model. The mean absolute error is inversely correlated with model performance. The measured coefficients are positively correlated with model performance.

[0079] In some embodiments, for at least two first fermentation product prediction models that can be applied to a digester, the performance indicators of the at least two first fermentation product prediction models can be determined based on a first test dataset.

[0080] The first test dataset corresponds to the digester. The first test dataset may include at least one first test data point. The first test data may include material parameter values, operating parameter values, and fermentation product values. The fermentation product values ​​can be understood as labels or actual values. The material parameters, operating parameters, and fermentation products are described in the preceding text. Specifically, the collected digester dataset can be split into a first sample dataset and a first test dataset. For example, the digester dataset can be randomly divided into an 80% first sample dataset and a 20% first test dataset. The digester dataset can be split directly. Alternatively, the digester data in the digester dataset can be preprocessed, and then the preprocessed digester dataset can be split. The preprocessing may include filtering and deleting digester data containing missing values.

[0081] In practical applications, the first test data in the first test dataset can be input into each trained first fermentation product prediction model to obtain the output of the first fermentation product prediction model; the performance index of the first fermentation product prediction model can be calculated based on the output of the first fermentation product prediction model and the true value of the first test data.

[0082] In some embodiments, for at least two first fermentation product prediction models that can be applied to leachate ponds, the performance indicators of the at least two first fermentation product prediction models can be determined based on a second test dataset.

[0083] The second test dataset corresponds to the leachate tank. The second test dataset may include at least one second test data set. The second test data set may include material parameter values, operating parameter values, and fermentation product values. The fermentation product values ​​can be understood as labels or actual values. The material parameters, operating parameters, and fermentation products are described in the preceding text. The collected leachate tank dataset can be split into a second sample dataset and a second test dataset. For example, the leachate tank dataset can be randomly divided into an 80% second sample dataset and a 20% second test dataset. The leachate tank dataset can be split directly. Alternatively, the leachate tank data in the leachate tank dataset can be preprocessed, and the preprocessed leachate tank dataset can be split. The preprocessing includes filtering and deleting leachate tank data containing missing values.

[0084] In practical applications, the second test data in the second test dataset can be input into each trained first fermentation product prediction model to obtain the output of the first fermentation product prediction model; the performance index of the first fermentation product prediction model can be calculated based on the output of the first fermentation product prediction model and the true value of the second test data.

[0085] Step 13: Select the target fermentation product prediction model from at least two first fermentation product prediction models based on performance indicators.

[0086] In some embodiments, for at least two first fermentation product prediction models that can be applied to a digester, a target fermentation product prediction model can be selected from the at least two first fermentation product prediction models based on performance indicators.

[0087] For example, the performance index may include mean absolute error. Then, the first fermentation product prediction model with the smallest mean absolute error can be selected from the at least two first fermentation product prediction models as the target fermentation product prediction model. Alternatively, the performance index may include measured coefficients. Then, the first fermentation product prediction model with the largest measured coefficients can be selected from the at least two first fermentation product prediction models as the target fermentation product prediction model. Furthermore, the performance index may include both mean absolute error and measured coefficients. Then, the first fermentation product prediction model with both the smallest mean absolute error and the largest measured coefficients can be selected from the at least two first fermentation product prediction models as the target fermentation product prediction model.

[0088] In some embodiments, for at least two first fermentation product prediction models that can be applied to leachate ponds, a target fermentation product prediction model can be selected from the at least two first fermentation product prediction models based on performance indicators.

[0089] For example, the performance index may include mean absolute error. Then, the first fermentation product prediction model with the smallest mean absolute error can be selected from the at least two first fermentation product prediction models as the target fermentation product prediction model. Alternatively, the performance index may include measured coefficients. Then, the first fermentation product prediction model with the largest measured coefficients can be selected from the at least two first fermentation product prediction models as the target fermentation product prediction model. Furthermore, the performance index may include both mean absolute error and measured coefficients. Then, the first fermentation product prediction model with both the smallest mean absolute error and the largest measured coefficients can be selected from the at least two first fermentation product prediction models as the target fermentation product prediction model.

[0090] Step 14: Predict the fermentation products based on the target fermentation product prediction model.

[0091] In some embodiments, step 13 yields a target fermentation product prediction model corresponding to the digester. Thus, at least two fermentation products from the digester can be predicted based on this model. In practical applications, the material and operating parameters of the digester can be input into the target fermentation product prediction model to obtain predicted values ​​for at least two fermentation products. The fermentation products include gases. The at least two gaseous fermentation products may include biogas and methane.

[0092] In some embodiments, step 13 yields a target fermentation product prediction model corresponding to the leachate tank. Thus, at least two fermentation products from the leachate tank can be predicted based on this model. In practical applications, the material and operational parameters of the leachate tank can be input into the target fermentation product prediction model to obtain predicted values ​​for at least two fermentation products. The fermentation products include gases. The at least two gaseous fermentation products include biogas and methane.

[0093] In some embodiments, an early warning function can also be implemented during the dry fermentation process.

[0094] After step 14, predicted biogas and methane values ​​for the digester can be obtained. The predicted biogas value includes biogas production. The predicted methane value includes methane content. Biogas and methane thresholds for the digester can be preset. Thus, the predicted biogas value can be compared with the biogas threshold to provide an early warning regarding biogas production; the predicted methane value can be compared with the methane threshold to provide an early warning regarding methane content. If the predicted biogas value is less than the biogas threshold, a first early warning message can be generated. The first early warning message indicates that the biogas production of the digester is lower than normal. If the predicted biogas value is greater than the biogas threshold, no warning is issued. If the predicted methane value is less than the methane threshold, a second early warning message can be generated. The second early warning message indicates that the methane content of the digester is lower than normal. If the predicted methane value is greater than the methane threshold, no warning is issued.

[0095] After step 14, the predicted biogas and methane values ​​for the leachate tank can be obtained. The predicted biogas value includes biogas production. The predicted methane value includes methane content. Biogas and methane thresholds for the leachate tank can be preset. Thus, the predicted biogas value can be compared with the biogas threshold to provide an early warning for biogas production; the predicted methane value can be compared with the methane threshold to provide an early warning for methane content. If the predicted biogas value is less than the biogas threshold, a third early warning message can be generated. This third early warning message indicates that the biogas production of the leachate tank is lower than the normal production. If the predicted biogas value is greater than the biogas threshold, no warning is issued. If the predicted methane value is less than the methane threshold, a fourth early warning message can be generated. This fourth early warning message indicates that the methane content of the leachate tank is lower than the normal content. If the predicted methane value is greater than the methane threshold, no warning is issued.

[0096] In some embodiments, guidance can also be provided for the control of operating parameters during dry fermentation.

[0097] The values ​​of the operating parameters can be determined based on the model construction method corresponding to the prediction model of the target fermentation product.

[0098] Specifically, in dry fermentation, material parameters are relatively objective and fixed. Operators often adjust these parameters to maximize the yield of fermentation products. Therefore, a second fermentation product prediction model can be constructed based on the model construction method corresponding to the target fermentation product prediction model. This second model is used to predict fermentation products based on the operating parameters. Since the second model can predict fermentation products solely based on the operating parameters without considering material parameters, the values ​​of the operating parameters can be determined using a particle swarm optimization algorithm. This facilitates operators in adjusting operating parameters to maximize the yield of fermentation products.

[0099] The at least two model building methods may include the AutoGluon method and the H2O method. Thus, the model building method corresponding to the target fermentation product prediction model can be either the AutoGluon method or the H2O method. In some scenario examples, for a garage-type digester, the R-value of the first fermentation product prediction model built based on the AutoGluon method relative to biogas production... 2 =0.74, MAE=60.38, and R relative to methane content 2 =0.83, MAE=3.21. For leachate ponds, the R-value of the first fermentation product prediction model built based on the AutoGluon method relative to biogas production is... 2 =0.88, MAE=61.03, and R relative to methane content 2 =0.89, MAE = 2.90. Therefore, the first fermentation product prediction model built based on the AutoGluon method can be selected as the target fermentation product prediction model. The model construction method corresponding to the target fermentation product prediction model can be the AutoGluon method. Compared with the H2O method, the target fermentation product prediction model based on the AutoGluon method has higher performance. The AutoGluon method, by integrating a large number of sub-models, can extremely accurately capture the relationship between input and output variables in the digester and leachate tank. Based on the AutoGluon method, the biogas production and methane content of the digester and leachate tank can be predicted stably and quickly, thus providing prediction and guidance for industrial production.

[0100] Please see Figure 3 The determination of operational parameter values ​​using the particle swarm optimization algorithm can specifically include the following steps.

[0101] Step 21: Determine the fitness of particles in the particle swarm based on the second fermentation product prediction model. The particles have position and velocity. The position represents the value of the operating parameter, and the velocity represents the degree of change of the value of the operating parameter.

[0102] The particle swarm can include multiple particles. The number of particles can be flexibly set as needed.

[0103] Each particle in the particle swarm can have a position. The position represents the value of an operational parameter. For example, the position can include a position vector, where the number of data elements can equal the number of operational parameters. Each data element in the position vector can correspond to an operational parameter, specifically a value of that operational parameter. Thus, the position of each particle can be understood as a combination of values ​​for multiple operational parameters during dry fermentation. The positions of multiple particles can be understood as multiple combinations of values ​​for multiple operational parameters during dry fermentation.

[0104] Each particle in the particle swarm may also possess a velocity. The velocity represents the degree of change in the value of the operating parameter, which may include the rate of change of the operating parameter value. For example, the velocity may include a velocity vector, where the number of data elements can be equal to the number of operating parameters. Each data element in the velocity vector may correspond to an operating parameter, representing the rate of change of a particular value of that operating parameter. Specifically, this particular value may include the data element in the particle's position vector corresponding to that operating parameter.

[0105] The fitness is used to represent the superiority or inferiority of a particle's position. The fitness value can be positively correlated with the degree of superiority or inferiority. The second fermentation product prediction model can be used as the objective function of a particle swarm optimization algorithm, and this objective function is used to determine the particle's fitness. Specifically, the particle's position can be input into the second fermentation product prediction model, and the fitness can be determined based on the model's output. For example, the output of the second fermentation product prediction model can be used as the fitness.

[0106] This involves collecting an optimized dataset. The optimized dataset may include at least one optimized data point. The optimized data may include operating parameter values ​​and fermentation product values. The fermentation product values ​​can be understood as labels for the optimized data. Therefore, a second fermentation product prediction model can be trained using the optimized dataset. For example, the optimized data from the optimized dataset can be input into the second fermentation product prediction model to obtain its output; loss information can be calculated using a loss function based on the output of the second fermentation product prediction model and the labels of the optimized data; and the parameters of the second fermentation product prediction model can be adjusted based on the loss information. The trained second fermentation product prediction model can be used to determine the fitness of particles.

[0107] Step 22: Update the individual optimal position of the particles and the global optimal position of the particle swarm based on fitness.

[0108] The values ​​of operational parameters are determined using the particle swarm optimization algorithm, which can include one or more iterations. Each iteration records the individual optimal position of a particle and the global optimal position of the particle swarm. Each particle has at least one historical position prior to the current iteration. The individual optimal position of a particle includes its historical position with the best fitness prior to the current iteration. The global optimal position of the particle swarm includes the optimal position of the individual particle with the best fitness in the swarm. It is worth noting that in the first iteration, the particle's position and velocity can be preset, for example, randomly set. In subsequent iterations, the particle's position and velocity can be updated from the previous iteration.

[0109] In the current iteration, each particle in the particle swarm can have a position and an individual optimal position. The individual optimal position includes the historical position of the particle with the best fitness before the current iteration. The particle swarm can also have a global optimal position. The global optimal position includes the individual optimal position of the particle swarm with the best fitness. The individual optimal position of each particle can be updated based on fitness. Specifically, for each particle in the particle swarm, the particle's fitness can be compared with the fitness of its individual optimal position; if the particle's fitness is better than the fitness of its individual optimal position, then the particle's position can be taken as the new individual optimal position; if the particle's fitness is worse than the fitness of its individual optimal position, then the individual optimal position can remain unchanged. After updating the individual optimal positions of each particle in the particle swarm, the global optimal position of the particle swarm can be updated based on the updated individual optimal fitness of each particle. Specifically, the optimal position of an individual with the best fitness can be selected from the individual optimal fitness of each particle; the fitness of the selected optimal position can be compared with the fitness of the global optimal position; if the fitness of the selected optimal position is better than the fitness of the global optimal position, the selected optimal position can be taken as the new global optimal position; if the fitness of the selected optimal position is worse than the fitness of the global optimal position, the global optimal position can be kept unchanged.

[0110] Step 23: Update the particle's position and velocity based on the individual optimal position and the global optimal position.

[0111] The positions and velocities of each particle in the particle swarm can be further updated based on the updated individual optimal positions and the updated global optimal positions. For example, the velocity of each particle can be updated using the following formula based on the updated individual optimal positions and the updated global optimal positions: k represents the current iteration number, and k+1 represents the next iteration number. This represents the updated velocity of the i-th particle. Let w represent the velocity of the i-th particle, and w represent the inertia factor. Gbest represents the optimal position of an individual. k c1 and c2 represent the global optimal position, and c1 and c2 represent the learning factors. and This represents a random number. The position of each particle can be updated based on the updated velocity using the following formula: k represents the current iteration number, and k+1 represents the next iteration number. This represents the updated velocity of the i-th particle. This represents the position of the i-th particle. This represents the updated position of the i-th particle.

[0112] Step 24: After the iteration is complete, the values ​​of the operation parameters can be determined based on the global optimal position.

[0113] Steps 21-23 can be executed iteratively until a set condition is met. This set condition may include reaching a set number of iterations. After the set condition is met, step S24 can be executed. The globally optimal position can be understood as a combination of values ​​for multiple operating parameters during dry fermentation. Therefore, the globally optimal position can be used as an optimized value for the operating parameters.

[0114] In some embodiments of this example, the operating parameters during the dry fermentation process can be set to the optimized values. By adjusting the operating parameters of the industrial-scale garage-type dry fermentation system, the anaerobic environment of the reactor can be improved, providing favorable conditions for microbial survival, promoting efficient reaction and product output, and facilitating the attainment of maximum biogas and methane yields.

[0115] In some embodiments of this example, the values ​​of the operating parameters determined by the particle swarm optimization algorithm can also be input into the second fermentation product prediction model to predict the maximum biogas yield and the maximum methane content.

[0116] In some embodiments of this example, particle swarm optimization (PSO) can be used to determine the values ​​of the digester's operating parameters. These parameters may include feed mass, pressure, temperature, spraying time, and number of sprays.

[0117] A first optimized dataset can be collected for the digester. This first optimized dataset may include at least one first optimized data point. The first optimized data may include operating parameter values ​​and fermentation product values. The operating parameter values ​​may include feedstock quality, pressure, temperature, spraying time, and spraying frequency. The fermentation products may include biogas and methane. The fermentation product values ​​may include biogas yield and methane content, which can be used as labels for the first optimized data. A second fermentation product prediction model can be trained using the first optimized dataset. Based on the trained second fermentation product prediction model corresponding to the digester, the values ​​of the digester's operating parameters can be determined using a particle swarm optimization algorithm.

[0118] In this way, MOPSO can be used to find the optimal combination of operating parameters in the digester, and then the system can explore ways to improve the digester's biogas and methane production performance with the highest number of iterations. The final output is the optimized operating parameter values, biogas yield, and methane content. In other words, by learning from existing digester data, the system iteratively finds the optimal values ​​of operating parameters that bring better performance, and uses these optimal values ​​to predict biogas yield and methane content.

[0119] The formulas involved in the process are as follows.

[0120] AutoML:Biogas Production and CH4=F AutoML (Inputs);

[0121] multi objectives:max.(Biogas Production+CH4)=max.F AutoML (Inputs);

[0122] constraints:LB of inputs≤values ​​of inputs≤UB of inputs;

[0123] Among them, F AutoML represents the optimal automated machine learning model connecting the input and output variables; Inputs represents the optimization conditions for each reactor; Biogas production and CH4 have equal weights in the multi-objective optimization; LB and UB represent the bottom and upper bounds of the search input boundary, respectively.

[0124] In some embodiments of this example, particle swarm optimization can be used to determine the values ​​of operating parameters for the leachate tank. These operating parameters may include temperature, pressure, liquid level, pH value, etc.

[0125] A second optimized dataset corresponding to the leachate tank can be collected. This second optimized dataset may include at least one second optimized data point. The second optimized data may include operating parameter values ​​and fermentation product values. The operating parameter values ​​may include temperature, pressure, pH, liquid level, etc. The fermentation products may include biogas and methane. The fermentation product values ​​may include biogas yield and methane content, which can be used as labels for the second optimized data. A second fermentation product prediction model can be trained using the second optimized dataset. Based on the trained second fermentation product prediction model corresponding to the leachate tank, the values ​​of the leachate tank's operating parameters can be determined using a particle swarm optimization algorithm.

[0126] In this way, MOPSO can be used to find the optimal combination of operating parameters in the leachate tank, and then the system can explore ways to improve the biogas and methane production performance of the leachate tank with the highest number of iterations. The final output is the optimized values ​​of the operating parameters, biogas yield, and methane content. That is, by learning from existing leachate tank data, the system iteratively finds the optimal values ​​of operating parameters that bring better performance, and uses these optimal values ​​to predict biogas yield and methane content.

[0127] The formulas involved in the process are as follows.

[0128] AutoML:Biogas Production and CH4=F AutoML (Inputs);

[0129] multi objectives:max.(Biogas Production+CH4)=max.F AutoML (Inputs);

[0130] constraints:LB of inputs≤values ​​of inputs≤UB of inputs;

[0131] Among them, F AutoML represents the optimal automated machine learning model connecting the input and output variables; Inputs represents the optimization conditions for each reactor; Biogas production and CH4 have equal weights in the multi-objective optimization; LB and UB represent the bottom and upper bounds of the search input boundary, respectively.

[0132] In some embodiments of this example, the predicted value of the fermentation product can be obtained after step 14. This allows for the analysis of the importance of multiple operating parameters to the predicted value of the fermentation product, enabling the selection of one or more operating parameters from among them. The values ​​of these one or more operating parameters can then be determined using a particle swarm optimization algorithm. This reduces the number of operating parameters whose values ​​need to be determined. Furthermore, since the number of data elements in particle position and particle velocity can be equal to the number of selected operating parameters, reducing the number of data elements in particle position and particle velocity reduces the computational load of the particle swarm optimization algorithm.

[0133] The importance of multiple operating parameters to the predicted fermentation product values ​​can be analyzed using methods such as SHAP value analysis and Pearson correlation coefficient analysis. These multiple operating parameters may include the parameters used in step 14 for predicting fermentation products. For example, multiple operating parameters for the digester may include feed mass, temperature, pressure, spraying time, and spraying frequency. Similarly, multiple operating parameters for the leachate tank may include temperature, pressure, liquid level, and pH.

[0134] Importance analysis can determine the degree of importance of multiple operational parameters to the predicted values ​​of fermentation products. One or more operational parameters with higher importance can be selected from these parameters. For example, importance analysis can obtain the importance index of multiple operational parameters to the predicted values ​​of fermentation products. One or more operational parameters with the highest importance index can be selected from these parameters. For instance, operational parameters for a digester may include feed quality, temperature, pressure, spraying time, and spraying frequency. Importance analysis can be used to select operational parameters such as spraying time, spraying frequency, and pressure. Using particle swarm optimization, a spraying time of 10 minutes / day and a pressure of 1000 Pa can be determined. Similarly, operational parameters for a leachate tank may include temperature, pressure, liquid level, and pH value. Importance analysis can be used to select operational parameters such as liquid level, pH, and total inorganic carbon. Using particle swarm optimization, a liquid level of 1500 mm, a pH range of 8.05-8.10, and a total inorganic carbon content greater than 13.8 g / L can be determined.

[0135] The fermentation data processing method for dry fermentation described in this specification can train a first fermentation product prediction model constructed based on at least two model building methods. This first fermentation product prediction model is used to predict fermentation products based on material parameters and operational parameters during the dry fermentation process. The method can determine the performance indicators of the at least two trained first fermentation product prediction models. A target fermentation product prediction model can be selected from the at least two first fermentation product prediction models based on the performance indicators. The fermentation products can then be predicted based on the target fermentation product prediction model. This method enables the prediction of fermentation products during the dry fermentation process, which is of great significance for guiding the safe operation and efficient production of a factory.

[0136] In addition, based on automated machine learning and metaheuristic algorithms, the relationship between operating parameters, reaction conditions and product properties can be better studied, thereby realizing multi-objective prediction and early warning guidance of biogas performance in industrial-scale garage-style dry fermentation.

[0137] Furthermore, this specification provides a novel intelligent method for multi-objective prediction and early warning guidance of biogas performance in industrial-scale garage-type dry fermentation based on automated machine learning. This novel intelligent method fully considers the impact of operating parameters on biogas yield and methane content, improving prediction accuracy and expanding its application scope.

[0138] The following example, using the digester of an industrial-scale garage-type dry fermentation system, illustrates a scenario described in this manual.

[0139] The digester pool dataset contains 620 data points. Models based on the AutoGluon and H2O methods were used to simulate and predict the digester pool dataset over a relatively long runtime (60-1200 s). Please refer to [link / reference]. Figures 4A-4D The model based on the AutoGluon method achieved optimal prediction performance after 60 seconds of operation. Regarding the R-value of biogas production in the digester... 2 =0.74, MAE=60.38. R for methane content in the digester 2 =0.83, MAE=3.21. The model based on the H2O method achieved its optimal prediction performance after 900s. The model based on the AutoGluon method performed significantly better than the model based on the H2O method, and its running time was shorter.

[0140] Figure 4A The variation in prediction performance of the model based on the AutoGluon method is shown over a runtime range of 60–1200 s. Figure 4B The variation in prediction performance of the H2O-based model over a runtime range of 60–1200 s is shown. Figure 4C The predictive performance of the model based on the AutoGluon method for biogas production in digesters is shown. Figure 4D The predictive performance of the AutoGluon-based model for methane content in the digester is shown. Figure 4C The top shading represents the distribution of predicted biogas production across the 95% confidence interval of the regression line. Figure 4C The shaded area on the right represents the distribution of actual biogas production across the 95% confidence interval of the regression line. Figure 4D The top shading represents the distribution of predicted methane content across the 95% confidence interval of the regression line. Figure 4D The shaded area on the right represents the distribution of the actual methane content across the 95% confidence interval of the regression line.

[0141] The AutoGluon method integrates multiple sub-models into a WeightedEnsemble model. As shown in Table 3, at a runtime of 60 s, this hybrid model has the lowest MAE (Maximum Effectiveness) for biogas production and methane content compared to individual sub-models, at 63.55 and 4.37, respectively. Among the multiple sub-models of the AutoGluon method, LightGBM performs best for biogas production prediction, with a validation MAE of 66.31; ExtraTrees performs best for methane content prediction, with a validation MAE of 4.54.

[0142] Please see Figures 5A-5H SHAP value analysis revealed that spraying time, spray frequency, and pressure are the most important operating parameters for the digester. A spraying strategy of 60m... 3 A self-priming pump with an hourly spraying time of 5 minutes per hour, a spraying time of 10 minutes per day, a pressure of 1000Pa, and sufficient feed are ensured to maximize the biogas production and methane content of the digester.

[0143] Figure 5A The results of the SHAP value analysis of biogas production from the digester are shown. Figure 5B The results of the SHAP value analysis of methane content in the digester are shown. Figure 5A and Figure 5B The right side indicates the order of importance of operating conditions. Figure 5C shows the SHAP value analysis results for digester temperature and biogas production. Figure 5D The results of the SHAP value analysis for spraying time and biogas production are shown. Figure 5E The results of the SHAP value analysis of digester pressure and biogas production are shown. Figure 5F The results of the SHAP value analysis for raw material quality and methane content are shown. Figure 5G The results of the SHAP value analysis for digester temperature and methane content are shown. Figure 5H The results of the SHAP value analysis of the total solids content and methane content of the raw materials are shown.

[0144] Table 3

[0145]

[0146]

[0147]

[0148] The following example, using the leachate tank of an industrial-scale garage-type dry fermentation system, illustrates another scenario in this manual.

[0149] The leachate treatment tank dataset contains 341 data points. Models using the AutoGluon and H2O methods were used to simulate and predict the leachate treatment tank dataset over a relatively long runtime (60-1200 s). Please refer to [link / reference]. Figures 6A-6D The model based on the AutoGluon method achieved optimal prediction performance after running for 300 seconds. Regarding the R-value of biogas production in leachate ponds... 2 =0.88, MAE=61.03. Regarding the R value for methane content in the leachate tank... 2 =0.89, MAE=2.90. The model based on the H2O method achieved its optimal prediction performance after 600s. The model based on the AutoGluon method performed significantly better than the model based on the H2O method, and its running time was shorter.

[0150] Figure 6A The variation in prediction performance of the model based on the AutoGluon method is shown over a runtime range of 60–1200 s. Figure 6B The variation in prediction performance of the H2O-based model over a runtime range of 60–1200 s is shown. Figure 6C The predictive performance of the model based on the AutoGluon method for biogas production in leachate ponds is shown. Figure 6D The predictive performance of the model based on the AutoGluon method for methane content in leachate ponds is shown. Figure 6C The top shading represents the distribution of predicted biogas production across the 95% confidence interval of the regression line. Figure 6C The shaded area on the right represents the distribution of actual biogas production across the 95% confidence interval of the regression line. Figure 6D The top shading represents the distribution of predicted methane content across the 95% confidence interval of the regression line. Figure 6D The shaded area on the right represents the distribution of the actual methane content across the 95% confidence interval of the regression line.

[0151] The AutoGluon method integrates multiple sub-models into a WeightedEnsemble model. As shown in Table 4, the WeightedEnsemble model has the lowest MAE (61.03 and 2.90) compared to the individual sub-models for biogas production and methane content prediction, respectively. Among the multiple sub-models of the AutoGluon method, CatBoost performs best for both biogas production and methane content prediction, with validation MAEs of 68.09 and 3.61, respectively.

[0152] Please see Figures 7A-7H SHAP value analysis revealed that liquid level, pH, and total inorganic carbon are the most important operating parameters for leachate tanks. Maintaining the liquid level at approximately 1500 mm, the optimal pH range of 8.05-8.10, a total inorganic carbon content greater than 13.8 g / L, and a ratio of volatile fatty acids to total inorganic carbon less than 0.3 can maximize the biogas production and methane content of the leachate tank.

[0153] Figure 7A The results of the SHAP value analysis of biogas production from the leachate pond are shown. Figure 7B The results of the SHAP value analysis of methane content in the leachate tank are shown. Figure 7A and Figure 7B The right side indicates the order of importance of the operating conditions. Figure 7C The results of the SHAP value analysis of leachate tank temperature and biogas production are shown. Figure 7D The results of the SHAP value analysis for liquid level and biogas production are shown. Figure 7E The results of the SHAP value analysis for pH and biogas production are shown. Figure 7F The results of the SHAP value analysis for leachate tank temperature and methane content are shown. Figure 7G The results of the SHAP value analysis for liquid level and methane content are shown. Figure 7H The results of the SHAP value analysis of total inorganic carbon and methane content are shown.

[0154] Table 4

[0155]

[0156]

[0157]

[0158] The following is another scenario example from this manual.

[0159] The practical application of the dry fermentation system was validated. As shown in Table 5, 15 data points were collected from the digester and leachate tank respectively for practical prediction and early warning. Please refer to [link / reference needed]. Figures 8A-8DThe actual prediction errors for biogas production and methane content in garage-type digesters were 14.00% and 4.93%, respectively, while those in leachate digesters were 7.81% and 5.40%, respectively. When the minimum thresholds for biogas production and methane content in garage-type digesters were set to 294 m³, the prediction errors were... 3 When the concentration is 50%, the minimum thresholds for biogas production and methane content in the leachate tank are set to 589 m³. 3 At 50%, the dry fermentation system can accurately detect abnormal production data below the minimum threshold in the test set with 100% accuracy. Furthermore, multi-objective particle swarm optimization was performed on the operating parameters of the garage-type digester and leachate tank based on optimization data from the digester and leachate tank, respectively.

[0160] Figure 8A The test results of biogas production in the digester are shown; Figure 8B The test results for methane content in the digester are shown; Figure 8C The test results of biogas production from the leachate pond are shown; Figure 8D The test results for the methane content in the leachate tank are shown.

[0161] Table 5

[0162]

[0163] Please see Figure 9 This specification also provides an embodiment of a fermentation data processing device for dry fermentation, comprising:

[0164] Training unit 91 is used to train a first fermentation product prediction model constructed based on at least two model construction methods. The first fermentation product prediction model is used to predict fermentation products based on material parameters and operating parameters of dry fermentation.

[0165] Determining unit 92 is used to determine the performance metrics of at least two first fermentation product prediction models after training.

[0166] Selecting unit 93 is used to select a target fermentation product prediction model from at least two first fermentation product prediction models based on performance indicators.

[0167] Prediction unit 94 is used to predict fermentation products based on the target fermentation product prediction model.

[0168] This invention provides a computer device, including a memory and a processor. The memory stores a computer program that can run on the processor, and the processor executes the computer program to implement… Figure 1 The corresponding implementation examples.

[0169] It should be noted that the various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, the device embodiments, computer device embodiments, and computer storage medium embodiments are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments. Furthermore, it is understood that those skilled in the art, after reading this specification, can arbitrarily combine some or all of the embodiments listed in this specification without creative effort, and such combinations are also within the scope of disclosure and protection of this specification.

[0170] Although this specification has been described by way of examples, those skilled in the art will recognize that many variations and modifications are possible with respect to this specification, and it is intended that the appended claims cover such variations and modifications without departing from the spirit of this specification.

Claims

1. A method for processing fermentation data during dry fermentation, characterized in that, include: A first fermentation product prediction model, constructed based on at least two model construction methods, is trained. This first fermentation product prediction model is used to predict fermentation products based on material parameters and operational parameters during the dry fermentation process. The operational parameters are those that can be adjusted by the operator during the dry fermentation process. The at least two model construction methods include the AutoGluon method and the H2O method. The first fermentation product prediction model constructed by each model construction method includes an ensemble model, which is obtained by integrating multiple sub-models. Determine the performance metrics of at least two first fermentation product prediction models after training. Select the target fermentation product prediction model from at least two first fermentation product prediction models based on performance indicators; Based on the target fermentation product prediction model, the fermentation product is predicted to obtain the predicted value of the fermentation product. The importance of multiple operating parameters to the predicted value of fermentation products is analyzed to obtain the importance index of multiple operating parameters to the predicted value of fermentation products. One or more operating parameters with the highest importance index are selected from the multiple operating parameters. Based on the model construction method corresponding to the target fermentation product prediction model, a second fermentation product prediction model is constructed. The second fermentation product prediction model is used to predict fermentation products based solely on operating parameters. Based on the second fermentation product prediction model, the values ​​of the selected operating parameters are determined by the particle swarm optimization algorithm; The values ​​of the operating parameters determined by the particle swarm optimization algorithm are input into the second fermentation product prediction model to predict the maximum biogas yield and the maximum methane content.

2. The method according to claim 1, characterized in that, The first fermentation product prediction model includes a multi-objective fermentation product prediction model.

3. The method according to claim 1, characterized in that, The steps for training the prediction model for the first fermentation product include: Based on the first sample dataset, a first fermentation product prediction model constructed using at least two model construction methods is trained. The first sample dataset corresponds to the digester and includes material parameter values, operating parameter values, and fermentation product values. Based on the second sample dataset, a first fermentation product prediction model constructed using at least two model construction methods is trained. The second sample dataset corresponds to the leachate tank and includes material parameter values, operating parameter values, and fermentation product values.

4. The method according to claim 1, characterized in that, The steps involved in selecting a prediction model for the target fermentation product include: Select the target fermentation product prediction model from at least two first fermentation product prediction models corresponding to the digester; Select the target fermentation product prediction model from at least two first fermentation product prediction models corresponding to the leachate tank; Accordingly, the steps for predicting fermentation products include: Based on the target fermentation product prediction model corresponding to the digester, at least two fermentation products of the digester are predicted. Based on the target fermentation product prediction model corresponding to the leachate tank, at least two fermentation products of the leachate tank are predicted.

5. The method according to claim 4, characterized in that, The fermentation products include gases, and at least two gaseous fermentation products include biogas and methane; the method further includes at least one of the following: The predicted biogas value of the digester is compared with the biogas threshold of the digester to provide early warning of biogas production in the digester. The predicted methane levels in the digester are compared with the methane threshold of the digester to provide an early warning of the methane content in the digester. The predicted biogas value of the leachate pond is compared with the biogas threshold of the leachate pond to provide early warning of biogas production in the leachate pond. The predicted methane levels in the leachate tank are compared with the methane threshold of the leachate tank to provide early warning of methane content in the leachate tank.

6. The method according to claim 1, characterized in that, The steps for determining the values ​​of the operation parameters include: The following steps are executed iteratively until the set conditions are met; Based on the second fermentation product prediction model, the fitness of particles in the particle swarm is determined. The particles have position and velocity, where position represents the value of the operating parameter and velocity represents the degree of change of the operating parameter value. Update the individual optimal position of the particle and the global optimal position of the particle swarm based on fitness. Update the particle's position and velocity based on the individual optimal position and the global optimal position; After the iteration is completed, the values ​​of the operation parameters are determined based on the global optimal position.

7. A fermentation data processing device for dry fermentation, characterized in that, include: A training unit is used to train a first fermentation product prediction model constructed based on at least two model construction methods. The first fermentation product prediction model is used to predict fermentation products based on material parameters and operating parameters of dry fermentation. The operating parameters are parameters that can be adjusted by the operator during the dry fermentation process. The at least two model construction methods include the AutoGluon method and the H2O method. The first fermentation product prediction model constructed by each model construction method includes an ensemble model, which is obtained by integrating multiple sub-models. A determination unit is used to determine the performance metrics of at least two first fermentation product prediction models after training. The selection unit is used to select a target fermentation product prediction model from at least two first fermentation product prediction models based on performance indicators; The prediction unit is used to predict the fermentation product based on the target fermentation product prediction model and obtain the predicted value of the fermentation product. The device is also used to analyze the importance of multiple operating parameters to the predicted value of fermentation products, obtain the importance index of multiple operating parameters to the predicted value of fermentation products, select one or more operating parameters with the highest importance index from multiple operating parameters; construct a second fermentation product prediction model according to the model construction method corresponding to the target fermentation product prediction model, the second fermentation product prediction model is used to predict fermentation products based solely on operating parameters; determine the values ​​of the selected operating parameters according to the second fermentation product prediction model using a particle swarm optimization algorithm; input the values ​​of the operating parameters determined by the particle swarm optimization algorithm into the second fermentation product prediction model to predict the maximum biogas yield and the maximum methane content.

8. A computer device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor executes the instructions to implement the method as described in any one of claims 1-6.

Images