Prediction and optimization method of microalgae flotation recovery efficiency based on machine learning

Abstract

The present invention relates to the technical field of microalgae biological solid-liquid separation, in particular to a method for predicting and optimizing the recovery efficiency of microalgae bead flotation based on machine learning. Different machine learning algorithms are introduced, and the multidimensional microalgae bead flotation data packets after screening are trained, calculated and screened based on different machine learning algorithms to obtain a trained ideal recovery efficiency prediction model; the established input features are predicted by the trained ideal recovery efficiency prediction model to obtain a predetermined recovery efficiency value, and the established recovery efficiency value is calculated using the SHAP library to determine the degree of influence of each established bead property and established operating conditions on the established recovery efficiency; the optimal bead flotation recovery conditions screened are arranged in descending order based on the degree of influence to enrich the data set of microalgae bead flotation, and the ideal recovery efficiency prediction model is optimized. The present invention can realize accurate prediction of the recovery efficiency of the microalgae bead flotation process based on the machine learning method, and significantly improve the recovery performance of the microalgae bead flotation.

Description

Technical Field

[0001] The present invention relates to the technical field of microalgae biological solid-liquid separation, and in particular to a method for predicting and optimizing microalgae floating bead flotation harvesting efficiency based on machine learning. Background Art

[0002] Microalgae are rich in valuable natural products such as proteins, lipids, carbohydrates, pigments, etc., and are an important source of biomass for the production of food, biofuels, cosmetics and medicines. In addition, microalgae play an important role in environmental protection. They can recover nutrients such as nitrogen and phosphorus from wastewater, effectively alleviating the problem of eutrophication of water bodies. As photosynthetic autotrophic organisms, microalgae can use solar energy for photosynthesis and absorb carbon dioxide, becoming a potential carbon capture and utilization technology with great potential for addressing climate change.

[0003] However, a major bottleneck in the microalgae industry is the biomass harvesting process, which usually accounts for 20% to 30% of the total production cost. This problem arises mainly due to the inherent characteristics of microalgae cultures, including low concentration, small cell size, negative surface charge, and cell density close to water. Currently, bead flotation technology, as an emerging microalgae harvesting method, has demonstrated its potential in microalgae harvesting. This technology is similar to traditional dissolved air flotation, but by using beads instead of microbubbles, the energy consumption of microbubble generation is significantly reduced.

[0004] At present, studies have found that the recovery efficiency of microalgae bead flotation is usually affected by a combination of multiple factors, and the emphasis on the individual contribution of each factor is limited. Experimental analysis usually requires a complete biochemical platform and professional technicians, and the process is time-consuming and expensive. In addition, it is impractical to analyze the impact of all variables on the recovery efficiency in a single experiment in terms of time, cost and feasibility. Therefore, it is difficult to establish a reliable connection between all influencing variables and recovery efficiency in microalgae bead flotation through experimental means, which leads to a major problem in practice for the target microalgae to use bead flotation to screen suitable operating conditions for efficient microalgae recovery, which hinders the development of microalgae bead flotation technology. Summary of the invention

[0005] The present invention overcomes the shortcomings of the prior art and provides a method for predicting and optimizing the flotation recovery efficiency of microalgae beads based on machine learning.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] The first aspect of the present invention provides a method for predicting and optimizing the recovery efficiency of microalgae float beads based on machine learning, comprising the following steps:

[0008] S102: obtaining a literature database on microalgae bead flotation, searching for microalgae characteristic data, bead property data, operation parameter data, and harvesting efficiency data of microalgae bead flotation through the literature database, and performing unit unification, missing value filling, and normalization on the microalgae characteristic data, bead property data, operation parameter data, and harvesting efficiency data, and then screening to obtain a screened multidimensional microalgae bead flotation data packet;

[0009] S104: Introducing different machine learning algorithms, training and calculating the screened multi-dimensional microalgae bead flotation data packets based on the different machine learning algorithms, obtaining the harvesting efficiency prediction model trained corresponding to each machine learning algorithm, and evaluating and optimizing each of the harvesting efficiency prediction models through multiple evaluation indicators and a grid search model, and performing optimal analysis and screening, to obtain the ideal harvesting efficiency prediction model trained;

[0010] S106: obtaining predetermined input features of microalgae bead flotation, importing the predetermined input features into the trained ideal recovery efficiency prediction model for prediction, obtaining a predetermined recovery efficiency value, using the SHAP library to calculate the predetermined recovery efficiency value to construct a SHAP Summary Plot visualization graph and analyze the distribution pattern of the SHAP value to determine the influence of each predetermined bead property and predetermined operating conditions on the predetermined recovery efficiency;

[0011] S108: Based on the descending order of the influence degree, the preferred predetermined input features with the highest ranking are obtained, an exhaustive method is introduced to change the characteristic values ​​of the preferred predetermined input features, and the optimal bead flotation harvesting conditions are predicted and screened out through the ideal harvesting efficiency prediction model, and the data set of microalgae bead flotation is enriched based on the optimal bead flotation harvesting conditions and the ideal harvesting efficiency prediction model is optimized.

[0012] More specifically, the step S102 includes the following steps:

[0013] Obtaining a literature database on microalgae bead flotation on the Web of Science platform, and searching for relevant data in microalgae bead flotation literature through the literature database to obtain microalgae characteristic data, bead property data, operation parameter data, and harvesting efficiency data of microalgae bead flotation;

[0014] Introducing a data unit conversion tool for the microalgae bead flotation experiment, using the data unit conversion tool to uniformly convert the microalgae characteristic data, bead property data, operating parameter data and harvesting efficiency data, and outputting the microalgae characteristic data, bead property data, operating parameter data and harvesting efficiency data after the uniform units;

[0015] Obtain the harvesting requirements of microalgae bead flotation, build a data value missing identification library according to the harvesting requirements, and use the data value missing identification library to perform numerical identification and fill in the microalgae characteristic data, bead property data, operation parameter data and harvesting efficiency data after unified units to fill the missing values;

[0016] After the filling is completed, the microalgae characteristic data, the bead property data, the operation parameter data and the harvesting efficiency data after the unified unit are processed by minimum-maximum normalization and a bundling operation is performed to obtain a multi-dimensional microalgae bead flotation data packet;

[0017] Preset a Spearman correlation coefficient threshold that affects the harvesting efficiency, calculate the Spearman correlation coefficient between each microalgae characteristic data, bead property data, and operation parameter data in the multi-dimensional microalgae bead flotation data packet and the harvesting efficiency, and obtain a plurality of Spearman correlation coefficients;

[0018] If the Spearman correlation coefficient is less than the Spearman correlation coefficient threshold, the data corresponding to the Spearman correlation coefficient is deleted from the multidimensional microalgae bead flotation data packet; if the Spearman correlation coefficient is greater than the Spearman correlation coefficient threshold, the data corresponding to the Spearman correlation coefficient is retained in the multidimensional microalgae bead flotation data packet, and finally the screened multidimensional microalgae bead flotation data packet is obtained.

[0019] More specifically, the step S104 includes the following steps:

[0020] The division ratio is set based on the harvesting demand of the microalgae bead flotation, and the screened multi-dimensional microalgae bead flotation data packet is randomly divided into a training set, a test set and a validation set according to the division ratio;

[0021] Introducing different machine learning algorithms, using the microalgae characteristic data, bead property data and operation parameter data in the screened multi-dimensional microalgae bead flotation data package as input features and the harvesting efficiency as output feature training rules, and performing model training on the training set in different machine learning algorithms based on the training rules, to obtain the harvesting efficiency prediction model completed by each machine learning algorithm;

[0022] Establishing an expected recovery efficiency prediction model according to the recovery requirements of the microalgae bead flotation, and determining multiple evaluation indicators of the expected recovery efficiency prediction performance and an ideal threshold value of each evaluation indicator according to the expected recovery efficiency prediction model;

[0023] The recovery efficiency prediction model trained by each machine learning algorithm is evaluated through multiple evaluation indicators of expected recovery efficiency prediction performance, and a grid search model is introduced simultaneously;

[0024] A grid search model is used to search for the optimal hyperparameters of the recovery efficiency prediction model trained corresponding to each machine learning algorithm, and an evaluation is performed based on multiple evaluation indicators and corresponding idealized thresholds to obtain the optimized evaluation indicator value of each recovery efficiency prediction model. The optimized evaluation indicator value is analyzed through radar map fitting and the optimal model is screened to obtain the trained ideal recovery efficiency prediction model.

[0025] More specifically, the grid search model is used to search for the optimal hyperparameters of the recovery efficiency prediction model trained corresponding to each machine learning algorithm, and the evaluation is performed based on multiple evaluation indicators and corresponding idealized thresholds to obtain the optimized evaluation indicator value of each recovery efficiency prediction model, and the optimized evaluation indicator value is analyzed by radar map fitting and the optimal model is screened to obtain the trained ideal recovery efficiency prediction model, which specifically includes the following steps:

[0026] During the evaluation process, the grid search model is used to search for the optimal hyperparameters of the recovery efficiency prediction model trained for each machine learning algorithm, so as to improve the recovery efficiency prediction capability of the optimized machine learning algorithm model according to the optimal hyperparameters. At this time, the optimized evaluation index values ​​of each evaluation index output by each trained recovery efficiency prediction model are obtained;

[0027] Constructing a radar map, fitting the idealized thresholds of the evaluation indicators into the radar map, and generating an idealized evaluation indicator radar map;

[0028] Fitting each of the trained recovery efficiency prediction models to output optimized evaluation index values ​​of each evaluation index in an idealized evaluation index radar map, and obtaining an optimized evaluation index radar map corresponding to each of the trained recovery efficiency prediction models;

[0029] Obtaining a radar area value of an idealized evaluation index radar map, defined as a first radar area value, and obtaining a radar area value of each optimized evaluation index radar map, defined as a second radar area value, and calculating the area difference between the second radar area value and the first radar area value one by one to obtain a plurality of area difference values;

[0030] Only the recovery efficiency prediction model corresponding to the trained recovery efficiency prediction model after optimization evaluation index radar map of the minimum area difference is extracted, and the extracted recovery efficiency prediction model after training is marked as the ideal recovery efficiency prediction model.

[0031] More specifically, the step S106 includes the following steps:

[0032] Obtaining predetermined microalgae characteristic data, predetermined float property data, and predetermined operation parameter data of microalgae float flotation, and bundling and marking them as predetermined input features;

[0033] The predetermined input characteristics are estimated and predicted by an ideal recovery efficiency prediction model to obtain a predetermined recovery efficiency value of microalgae bead flotation;

[0034] Based on the trained ideal recovery efficiency prediction model, a SHAP library is introduced into the big data network, and the SHAP between the predetermined bead properties and the predetermined operating conditions and the predetermined recovery efficiency value is calculated through the SHAP library to obtain multiple SHAP values;

[0035] Construct a SHAP Summary Plot graph based on the SHAP Summary Plot function, write the multiple SHAP values ​​into the SHAP Summary Plot graph for fitting, and obtain a SHAP Summary Plot visualization graph;

[0036] The distribution of multiple SHAP values ​​representing the relationship between the given bead properties and the given operating conditions and the given recovery efficiency value is extracted through the SHAP Summary Plot visualization diagram to obtain a SHAP value distribution pattern diagram. The degree of influence of each given bead property and given operating condition on the given recovery efficiency is determined based on the multiple SHAP value distribution patterns of the SHAP value distribution pattern diagram.

[0037] More specifically, the step S108 includes the following steps:

[0038] Arrange each predetermined bead property and the predetermined input characteristics corresponding to the predetermined operating conditions for the predetermined recovery efficiency in descending order based on the influence degree to obtain a descending order table of the predetermined input characteristics;

[0039] Preset the optimal input feature ranking limit, extract one or more predetermined input features above the optimal input feature ranking limit from the predetermined input feature descending sorting table, and mark them as preferred predetermined input features;

[0040] An exhaustive method is introduced to change the characteristic values ​​of the one or more preferred predetermined input characteristics to obtain the changed preferred predetermined input characteristics, and the one or more changed preferred predetermined input characteristics are predicted by the trained ideal recovery efficiency prediction model to obtain a new recovery efficiency value;

[0041] Only the bead properties and operating parameter values ​​under the maximum new recovery efficiency value are selected, and the bead properties and operating parameter values ​​under the maximum new recovery efficiency value are set as the optimal bead flotation recovery conditions;

[0042] Applying the optimal bead flotation recovery conditions to laboratory microalgae bead flotation related experiments to obtain an actual recovery efficiency value generated by the microalgae bead flotation related experiments, calculating the difference between the actual recovery efficiency value and the established recovery efficiency value, and obtaining a recovery efficiency deviation value;

[0043] If the recovery efficiency deviation value is greater than the preset recovery efficiency deviation value, the data set of microalgae bead flotation is enriched according to the recovery efficiency deviation value, and the prediction performance of the ideal recovery efficiency prediction model is optimized to improve the recovery efficiency.

[0044] The second aspect of the present invention provides a microalgae bead flotation recovery efficiency prediction and optimization system based on machine learning, the microalgae bead flotation recovery efficiency prediction and optimization system includes a memory and a processor, the memory stores a microalgae bead flotation recovery efficiency prediction and optimization method program based on machine learning, and when the microalgae bead flotation recovery efficiency prediction and optimization method program is executed by the processor, any one of the steps of the microalgae bead flotation recovery efficiency prediction and optimization method is implemented.

[0045] The present invention solves the technical defects existing in the background technology, and the beneficial technical effects of the present invention are:

[0046] The microalgae characteristic data, bead property data, operation parameter data and recovery efficiency data of the microalgae bead flotation are obtained and the data are screened after unit unification, missing value filling and normalization processing to obtain the screened multidimensional microalgae bead flotation data package; different machine learning algorithms are introduced, and the screened multidimensional microalgae bead flotation data package is trained and calculated based on different machine learning algorithms to obtain the recovery efficiency prediction model corresponding to each machine learning algorithm, evaluate and optimize each recovery efficiency prediction model and perform optimal analysis and screening to obtain the ideal recovery efficiency prediction model after training. Model; predict the given input features through the trained ideal recovery efficiency prediction model to obtain the given recovery efficiency value, use the SHAP library to calculate the SHAP value of the given recovery efficiency value to determine the influence of each given bead property and given operating conditions on the given recovery efficiency; obtain the preferred given input features with the highest ranking based on the influence degree in descending order, change the characteristic value of the preferred given input features to screen out the optimal bead flotation recovery conditions, enrich the data set of microalgae bead flotation based on the optimal bead flotation recovery conditions and optimize the ideal recovery efficiency prediction model. The present invention combines the cross-knowledge of the field of microalgae solid-liquid separation and the field of computer science, and overcomes the high investment in time and labor of traditional single factor and orthogonal experimental methods. By applying the machine learning algorithm, a relationship model between the various influencing factors and the recovery efficiency in the process of microalgae bead flotation is established, and the accurate prediction of the recovery efficiency under known conditions is realized. In addition, based on the prediction model, the bead flotation recovery efficiency and microalgae resource utilization can be optimized for specific target microalgae species, thereby improving the overall recovery effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, drawings of other embodiments can be obtained based on these drawings without paying creative work.

[0048] Figure 1 The process of predicting and optimizing the recovery efficiency of microalgae float flotation based on machine learning is shown. Figure 1 ;

[0049] Figure 2 The process of predicting and optimizing the recovery efficiency of microalgae float flotation based on machine learning is shown. Figure 2 ;

[0050] Figure 3 The system framework diagram of the microalgae bead flotation harvesting efficiency prediction and optimization system based on machine learning is shown;

[0051] Figure 4 It is a schematic diagram of the feature correlation matrix of the multi-dimensional microalgae bead flotation data package;

[0052] Figure 5 It is a schematic diagram of the prediction data based on the back propagation neural network prediction model;

[0053] Figure 6 The SHAP value distribution diagram of the influence of float properties and operating conditions on recovery efficiency. DETAILED DESCRIPTION

[0054] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0055] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0056] The first aspect of the present invention provides a method for predicting and optimizing the recovery efficiency of microalgae flotation beads based on machine learning, such as Figure 1 As shown, the following steps are included:

[0057] S102: obtaining a literature database on microalgae bead flotation, searching for microalgae characteristic data, bead property data, operation parameter data, and harvesting efficiency data of microalgae bead flotation through the literature database, and performing unit unification, missing value filling, and normalization on the microalgae characteristic data, bead property data, operation parameter data, and harvesting efficiency data, and then screening to obtain a screened multidimensional microalgae bead flotation data packet;

[0058] S104: Introducing different machine learning algorithms, training and calculating the screened multi-dimensional microalgae bead flotation data packets based on the different machine learning algorithms, obtaining the harvesting efficiency prediction model trained corresponding to each machine learning algorithm, and evaluating and optimizing each of the harvesting efficiency prediction models through multiple evaluation indicators and a grid search model, and performing optimal analysis and screening, to obtain the ideal harvesting efficiency prediction model trained;

[0059] S106: obtaining predetermined input features of microalgae bead flotation, importing the predetermined input features into the trained ideal recovery efficiency prediction model for prediction, obtaining a predetermined recovery efficiency value, using the SHAP library to calculate the predetermined recovery efficiency value to construct a SHAP Summary Plot visualization graph and analyze the distribution pattern of the SHAP value to determine the influence of each predetermined bead property and predetermined operating conditions on the predetermined recovery efficiency;

[0060] S108: Based on the descending order of the influence degree, the preferred predetermined input features with the highest ranking are obtained, an exhaustive method is introduced to change the characteristic values ​​of the preferred predetermined input features, and the optimal bead flotation harvesting conditions are predicted and screened out through the ideal harvesting efficiency prediction model, and the data set of microalgae bead flotation is enriched based on the optimal bead flotation harvesting conditions and the ideal harvesting efficiency prediction model is optimized.

[0061] More specifically, the step S102 is as follows: Figure 2 As shown, the specific steps include:

[0062] S202: obtaining a literature database on microalgae bead flotation on the Web of Science platform, and searching for relevant data in the literature on microalgae bead flotation through the literature database to obtain microalgae characteristic data, bead property data, operation parameter data, and harvesting efficiency data of microalgae bead flotation;

[0063] S204: introducing a data unit conversion tool for the microalgae bead flotation experiment, using the data unit conversion tool to convert the microalgae characteristic data, bead property data, operation parameter data and harvesting efficiency data into unified units, and outputting the microalgae characteristic data, bead property data, operation parameter data and harvesting efficiency data in unified units;

[0064] S206: Acquire the harvesting requirements of microalgae float flotation, build a data value missing identification library according to the harvesting requirements, and use the data value missing identification library to perform numerical recognition and fill in the microalgae characteristic data, float property data, operation parameter data and harvesting efficiency data after unified units to fill the missing values;

[0065] S208: After the filling is completed, the microalgae characteristic data, the bead property data, the operation parameter data and the harvesting efficiency data after the unified unit are processed by minimum-maximum normalization and a bundling operation is performed to obtain a multi-dimensional microalgae bead flotation data packet;

[0066] S210: Preset a Spearman correlation coefficient threshold that affects the harvesting efficiency, calculate the Spearman correlation coefficient between each microalgae characteristic data, bead property data, and operation parameter data in the multi-dimensional microalgae bead flotation data packet and the harvesting efficiency, and obtain a plurality of Spearman correlation coefficients;

[0067] S212: If the Spearman correlation coefficient is less than the Spearman correlation coefficient threshold, the data corresponding to the Spearman correlation coefficient is deleted from the multidimensional microalgae bead flotation data packet; if the Spearman correlation coefficient is greater than the Spearman correlation coefficient threshold, the data corresponding to the Spearman correlation coefficient is retained in the multidimensional microalgae bead flotation data packet, and finally a screened multidimensional microalgae bead flotation data packet is obtained.

[0068] It should be noted that there are a lot of literature data on microalgae bead flotation experiments, and these data are relatively messy, the units are not unified, and some relevant data are missing, which is not conducive to the accurate training and construction of the subsequent microalgae bead flotation recovery efficiency prediction model. Therefore, this method first collects the relevant sample data of microalgae bead flotation by searching the peer-reviewed literature published in the past 20 years on the Web of Science platform; the data unit conversion tool is used to convert the microalgae characteristic data, bead property data, operation parameter data, and recovery efficiency data of microalgae bead flotation retrieved from the literature database into a unified unit, which helps to eliminate the errors caused by different units and improve the comparability of the data. And according to the harvesting requirements of microalgae bead flotation, a data value missing identification library is constructed to fill the data, and the classification features are uniquely encoded, so that the microalgae characteristic data, bead property data, operation parameter data and harvesting efficiency data of microalgae bead flotation can comply with the formulated microalgae bead flotation harvesting target requirements to the greatest extent, ensure the integrity of the data, avoid the impact of data missing on the analysis results, provide a high-quality data foundation for subsequent analysis and modeling, improve data consistency and availability, reduce human errors, and better serve the optimization of harvesting efficiency of subsequent microalgae bead flotation experiments. Finally, the Spearman correlation coefficient is used to construct the feature correlation matrix (such as Figure 4 The input features that affect the recovery efficiency are screened out by using the feature correlation matrix (as shown in the figure), forming a feature correlation matrix, removing features with low correlation, and simplifying the data set to remove redundant or irrelevant features, which can effectively improve the efficiency and accuracy of subsequent modeling or analysis, reduce the complexity of the model, and reduce the model training time, so that the model is more concise and has better generalization ability. This method can reveal the key influencing factors in the flotation process of microalgae beads. On the one hand, it provides a high-quality data foundation for the analysis and modeling of subsequent prediction models, improves the consistency and availability of data, and reduces human errors. On the other hand, it provides a scientific basis for optimizing flotation technology and improving efficiency.

[0069] It should be noted that if the Spearman correlation coefficient is less than the Spearman correlation coefficient threshold, it means that the influence of this feature on the harvesting efficiency is less relevant, so it is deleted; otherwise, it means that the correlation is higher. Among them, the microalgae characteristics include microalgae species and microalgae density; the bead properties include bead type, bead size and bead concentration; the operating parameters include stirring speed, temperature, pH, flocculant, flocculant concentration and ion concentration. Among them, the algae species are Chlorella, Scenedesmus, and Spirulina, the microalgae density range is 0.15-0.85 mg / L, the bead types are sodium borosilicate microbeads and fly ash microbeads, the bead size is 6.68-95 µm, the bead concentration is 0.1-1.1 g / L, the stirring speed is 30-200 rpm, the temperature is 25-100 ℃, the flocculants are ferric chloride and chitosan, the flocculant concentration is 0-100 mg / L, and the ion concentration is 0-700 mM. The units of the same type of data were standardized, including microalgae density in mg / L, solution ion concentration in mM, stirring speed in rpm, and filled with room temperature 25°C and corresponding culture medium ion concentration.

[0070] More specifically, the step S104 includes the following steps:

[0071] The division ratio is set based on the harvesting demand of the microalgae bead flotation, and the screened multi-dimensional microalgae bead flotation data packet is randomly divided into a training set, a test set and a validation set according to the division ratio;

[0072] Introducing different machine learning algorithms, using the microalgae characteristic data, bead property data and operation parameter data in the screened multi-dimensional microalgae bead flotation data package as input features and the harvesting efficiency as output feature training rules, and performing model training on the training set in different machine learning algorithms based on the training rules, to obtain the harvesting efficiency prediction model completed by each machine learning algorithm;

[0073] Establishing an expected recovery efficiency prediction model according to the recovery requirements of the microalgae bead flotation, and determining multiple evaluation indicators of the expected recovery efficiency prediction performance and an ideal threshold value of each evaluation indicator according to the expected recovery efficiency prediction model;

[0074] The recovery efficiency prediction model trained by each machine learning algorithm is evaluated through multiple evaluation indicators of expected recovery efficiency prediction performance, and a grid search model is introduced simultaneously;

[0075] A grid search model is used to search for the optimal hyperparameters of the recovery efficiency prediction model trained corresponding to each machine learning algorithm, and an evaluation is performed based on multiple evaluation indicators and corresponding idealized thresholds to obtain the optimized evaluation indicator value of each recovery efficiency prediction model. The optimized evaluation indicator value is analyzed through radar map fitting and the optimal model is screened to obtain the trained ideal recovery efficiency prediction model.

[0076] It should be noted that due to the wide variety of machine learning algorithms, and the fact that the recovery efficiency prediction models trained by each machine learning algorithm have different degrees of performance differences, in order to ensure the stability and reliability of accurate prediction of the recovery efficiency in the process of microalgae bead flotation, and to further improve the recovery performance of microalgae bead flotation, it is necessary to train and construct an optimal recovery efficiency prediction model for application in microalgae bead flotation experiments. Among them, the division ratio can be divided into training set and test set according to the ratio of 70:30, 75:25 or 80:20. Machine learning algorithms include linear regression, Bayesian ridge regression, partial least squares regression, back propagation neural network, random forest, decision tree, support vector machine and K-nearest neighbor algorithm. Taking the prediction model example of back propagation neural network (such as Figure 5 As described above, the given known data is used to extract the required characteristic values ​​and pre-process them as the input variables of the above prediction model. After running, the recovery efficiency is 84.94%. Under the current laboratory conditions, the recovery efficiency obtained by flotation of microalgae with the given known data as the operating parameters is 86.6±2.8%. Through this method, different machine learning algorithms can be used to accurately train the screened multi-dimensional microalgae flotation data packets to obtain the recovery efficiency prediction model of the microalgae flotation process under different machine learning, and use several evaluation indicators and hyperparameter search to further evaluate and optimize these prediction models, so that a recovery efficiency prediction model representing the machine learning with the best prediction performance can be selected; by comparing the performance of multiple algorithms and models, the recovery efficiency prediction model is ensured to be accurate and stable in practical applications, so that the credibility of the recovery efficiency acquisition of the microalgae flotation process is greatly increased, and the optimization accuracy of subsequent microalgae flotation is guaranteed.

[0077] More specifically, the grid search model is used to search for the optimal hyperparameters of the recovery efficiency prediction model trained corresponding to each machine learning algorithm, and the evaluation is performed based on multiple evaluation indicators and corresponding idealized thresholds to obtain the optimized evaluation indicator value of each recovery efficiency prediction model, and the optimized evaluation indicator value is analyzed by radar map fitting and the optimal model is screened to obtain the trained ideal recovery efficiency prediction model, which specifically includes the following steps:

[0078] During the evaluation process, the grid search model is used to search for the optimal hyperparameters of the recovery efficiency prediction model trained for each machine learning algorithm, so as to improve the recovery efficiency prediction capability of the optimized machine learning algorithm model according to the optimal hyperparameters. At this time, the optimized evaluation index values ​​of each evaluation index output by each trained recovery efficiency prediction model are obtained;

[0079] Constructing a radar map, fitting the idealized thresholds of the evaluation indicators into the radar map, and generating an idealized evaluation indicator radar map;

[0080] Fitting each of the trained recovery efficiency prediction models to output optimized evaluation index values ​​of each evaluation index in an idealized evaluation index radar map, and obtaining an optimized evaluation index radar map corresponding to each of the trained recovery efficiency prediction models;

[0081] Obtaining a radar area value of an idealized evaluation index radar map, defined as a first radar area value, and obtaining a radar area value of each optimized evaluation index radar map, defined as a second radar area value, and calculating the area difference between the second radar area value and the first radar area value one by one to obtain a plurality of area difference values;

[0082] Only the recovery efficiency prediction model corresponding to the trained recovery efficiency prediction model after optimization evaluation index radar map of the minimum area difference is extracted, and the extracted recovery efficiency prediction model after training is marked as the ideal recovery efficiency prediction model.

[0083] It should be noted that for the evaluation and comparison of the recovery efficiency prediction model trained by each machine learning algorithm, the traditional method usually requires the calculation and comparison of each evaluation index for each prediction model, and the optimization process needs to rely on manual intervention to manually adjust the hyperparameters of the prediction model to achieve optimization, which greatly reduces the screening rate of the prediction model evaluation optimization between different machine learning algorithms, consumes a lot of manpower output costs, and there are large human intervention errors in the evaluation and optimization results, resulting in the prediction performance of the final selected recovery efficiency prediction model is unsatisfactory, which is not conducive to the accurate optimization application of recovery efficiency and data sets in the microalgae bead flotation process. In this regard, this method improves the prediction of recovery efficiency of the machine learning algorithm model by grid searching the optimal hyperparameters of the model, where the searched hyperparameters include L2 regularization parameter, Gamma distribution parameter, learning rate, number of neurons, number of trees, maximum depth of the tree, regularization parameter, kernel function and number of neighbors, so that the evaluation and comparison of the prediction model under each machine learning algorithm can achieve the maximum optimization state, reducing the phenomenon of large prediction performance errors during comparison and screening. This method fits the index values ​​of the optimized prediction models trained by each machine learning algorithm in the form of radar maps. The radar map can simultaneously display the evaluation values ​​and change ranges of various evaluation indicators on the same plane. Therefore, it is only necessary to compare the radar area differences between the radar maps corresponding to each prediction model to quickly know the pros and cons of the prediction performance evaluation between each model. Compared with the traditional manual comparison of each evaluation indicator one by one, the evaluation rate and evaluation accuracy are greatly improved. This method can make the prediction model used in the optimization of microalgae bead flotation harvest prediction more reliable and accurate, and improve the harvest quality of microalgae bead flotation.

[0084] More specifically, the step S106 includes the following steps:

[0085] Obtaining predetermined microalgae characteristic data, predetermined float property data, and predetermined operation parameter data of microalgae float flotation, and bundling and marking them as predetermined input features;

[0086] The predetermined input characteristics are estimated and predicted by an ideal recovery efficiency prediction model to obtain a predetermined recovery efficiency value of microalgae bead flotation;

[0087] Based on the trained ideal recovery efficiency prediction model, a SHAP library is introduced into the big data network, and the SHAP between the predetermined bead properties and the predetermined operating conditions and the predetermined recovery efficiency value is calculated through the SHAP library to obtain multiple SHAP values;

[0088] Construct a SHAP Summary Plot graph based on the SHAP Summary Plot function, write the multiple SHAP values ​​into the SHAP Summary Plot graph for fitting, and obtain a SHAP Summary Plot visualization graph;

[0089] The distribution of multiple SHAP values ​​representing the relationship between the given bead properties and the given operating conditions and the given recovery efficiency value is extracted through the SHAP Summary Plot visualization diagram to obtain a SHAP value distribution pattern diagram. The degree of influence of each given bead property and given operating condition on the given recovery efficiency is determined based on the multiple SHAP value distribution patterns of the SHAP value distribution pattern diagram.

[0090] It should be noted that the properties of the floats and operating conditions of microalgae float flotation have a crucial impact on the recovery efficiency. For example, too high a density of the floats may cause the floats to sink and reduce the recovery efficiency; for example, too high a concentration of microalgae may cause the surface of the floats to be oversaturated, which in turn reduces the efficiency. Therefore, in order to further improve the recovery efficiency and optimize the quality of microalgae float flotation, it is very necessary to calculate the degree of influence of the properties of the floats and operating conditions on the recovery efficiency. This method calculates the SHAP value by introducing the SHAP library, and then fits multiple SHAP values ​​with the SHAP Summary Plot function to display its distribution pattern. Since the features in the SHAP Summary Plot are sorted according to their importance to the prediction model. Important and influential features are at the top, and less influential features are at the bottom. Therefore, by looking at the SHAP value distribution of each feature, the relationship between the feature and the prediction result can be known; for example, if the SHAP value of a feature shows a positive distribution at a higher feature value, it means that the higher value of the feature helps to improve the recovery efficiency prediction result. Therefore, this method uses a distribution pattern diagram of multiple SHAP values ​​(such as Figure 6 As shown in the figure, the influence of each given bead property and given operating conditions on the given recovery efficiency can be quickly and accurately determined, providing a stable and reliable analysis basis for further improving the recovery efficiency and optimizing the quality of microalgae bead flotation, better improving the recovery efficiency of microalgae bead flotation, and making the optimization of microalgae bead flotation data more accurate. Among them, the Chinese name of the SHAP Summary Plot is the SHAP feature influence summary plot, which is used to comprehensively summarize the importance of all features in the recovery efficiency prediction model and their influence on the prediction results.

[0091] More specifically, the step S108 includes the following steps:

[0092] Arrange each predetermined bead property and the predetermined input characteristics corresponding to the predetermined operating conditions for the predetermined recovery efficiency in descending order based on the influence degree to obtain a descending order table of the predetermined input characteristics;

[0093] Preset the optimal input feature ranking limit, extract one or more predetermined input features above the optimal input feature ranking limit from the predetermined input feature descending sorting table, and mark them as preferred predetermined input features;

[0094] An exhaustive method is introduced to change the characteristic values ​​of the one or more preferred predetermined input characteristics to obtain the changed preferred predetermined input characteristics, and the one or more changed preferred predetermined input characteristics are predicted by the trained ideal recovery efficiency prediction model to obtain a new recovery efficiency value;

[0095] Only the bead properties and operating parameter values ​​under the maximum new recovery efficiency value are selected, and the bead properties and operating parameter values ​​under the maximum new recovery efficiency value are set as the optimal bead flotation recovery conditions;

[0096] Applying the optimal bead flotation recovery conditions to laboratory microalgae bead flotation related experiments to obtain an actual recovery efficiency value generated by the microalgae bead flotation related experiments, calculating the difference between the actual recovery efficiency value and the established recovery efficiency value, and obtaining a recovery efficiency deviation value;

[0097] If the recovery efficiency deviation value is greater than the preset recovery efficiency deviation value, the data set of microalgae bead flotation is enriched according to the recovery efficiency deviation value, and the prediction performance of the ideal recovery efficiency prediction model is optimized to improve the recovery efficiency.

[0098] It should be noted that, through this method, the trained ideal recovery efficiency prediction model can be used for a certain microalgae solution, and the input features can be ranked based on their influence on the recovery efficiency, so as to obtain the top-ranked input features to predict new recovery efficiency values, that is, one or more preferred predetermined input features selected by ranking are introduced into the ideal recovery efficiency prediction model to predict new recovery efficiency values ​​one by one, which can provide a favorable and reliable screening data basis for the selection of optimal microalgae bead flotation recovery conditions, effectively improve the accuracy of the microalgae bead flotation test, and make the microalgae bead flotation test tend to be optimal. Finally, the difference between the actual harvesting efficiency and the established harvesting efficiency is calculated and analyzed. If the harvesting efficiency deviation value is greater than the preset harvesting efficiency deviation value, it means that the error between the predicted result of the trained ideal harvesting efficiency prediction model and the harvesting efficiency output by the actual microalgae bead flotation test is large, which means that the model is extremely inaccurate. Therefore, it is necessary to enrich the data set of microalgae bead flotation and optimize the prediction performance of the ideal harvesting efficiency prediction model to improve its harvesting efficiency prediction accuracy. This method can further improve and optimize the harvesting efficiency prediction model, greatly enhance the prediction performance of the prediction model in the application process of microalgae bead flotation, and make the harvesting efficiency prediction results more accurate and reliable.

[0099] In addition, the following is some experimental data on how to use float beads to achieve efficient harvesting of Chlorella with a density of 0.25 mg / L:

[0100] (1) For the ideal recovery efficiency prediction model that was finally trained, the order of importance of the features that affect the recovery efficiency was obtained through SHAP analysis. The order was: microalgae density > ion concentration > flocculant concentration > bead size > bead concentration > pH > stirring speed > temperature.

[0101] (2) By setting the gradient of characteristic variables, the ion concentration range is 19-300 mM, with an interval of 5; the flocculant concentration range is 0-100 mg / L, with an interval of 5; the bead particle size range is 45-90 µm, with an interval of 5. Other parameters include bead concentration of 0.55 g / L, pH of 6, and stirring speed of 120 rpm. The exhaustive method was used to obtain the optimal operating parameter range, thereby achieving efficient flotation harvest of microalgae.

[0102] (3) According to the model prediction, it was found that when the ion concentration was 232-282 mM, the flocculant concentration was 50.55 mg / L, and the bead size was 55-90 µm, the recovery efficiency was higher than 85%. The optimal recovery conditions were when the ion concentration was 282 mM, the flocculant concentration was 50 mg / L, and the bead size was 90 µm. Other parameters were bead concentration 0.55 mg / L, pH 6, and stirring speed 120 rpm. The maximum recovery efficiency was 86.87%. Under laboratory conditions, the recovery efficiency of microalgae by bead flotation was 88.6±3.2%.

[0103] The second aspect of the present invention provides a microalgae bead flotation recovery efficiency prediction and optimization system based on machine learning, such as Figure 3 As shown, the microalgae bead flotation recovery efficiency prediction and optimization system includes a memory 31 and a processor 32. The memory 31 stores a microalgae bead flotation recovery efficiency prediction and optimization method program based on machine learning. When the microalgae bead flotation recovery efficiency prediction and optimization method program is executed by the processor 32, any one of the microalgae bead flotation recovery efficiency prediction and optimization method steps is implemented.

[0104] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting and optimizing the harvesting efficiency of microalgae float beads based on machine learning, characterized in that: The following steps are involved: S102: obtaining a literature database on microalgae bead flotation, searching for microalgae characteristic data, bead property data, operation parameter data, and harvesting efficiency data of microalgae bead flotation through the literature database, and performing unit unification, missing value filling, and normalization on the microalgae characteristic data, bead property data, operation parameter data, and harvesting efficiency data, and then screening to obtain a screened multidimensional microalgae bead flotation data packet; S104: Introducing different machine learning algorithms, training and calculating the screened multi-dimensional microalgae bead flotation data packets based on the different machine learning algorithms, obtaining the harvesting efficiency prediction model trained corresponding to each machine learning algorithm, and evaluating and optimizing each of the harvesting efficiency prediction models through multiple evaluation indicators and a grid search model, and performing optimal analysis and screening, to obtain the ideal harvesting efficiency prediction model trained; S106: obtaining predetermined input features of microalgae bead flotation, importing the predetermined input features into the trained ideal recovery efficiency prediction model for prediction, obtaining a predetermined recovery efficiency value, using the SHAP library to calculate the predetermined recovery efficiency value to construct a SHAP Summary Plot visualization graph and analyze the distribution pattern of the SHAP value to determine the influence of each predetermined bead property and predetermined operating conditions on the predetermined recovery efficiency; S108: based on the descending order of the influence degree, the preferred predetermined input features with the highest ranking are obtained, an exhaustive method is introduced to change the characteristic value of the preferred predetermined input features, and the optimal bead flotation recovery conditions are predicted and screened out through the ideal recovery efficiency prediction model, and the data set of microalgae bead flotation is enriched based on the optimal bead flotation recovery conditions and the ideal recovery efficiency prediction model is optimized; The step S104 specifically includes the following steps: The division ratio is set based on the harvesting demand of the microalgae bead flotation, and the screened multi-dimensional microalgae bead flotation data packet is randomly divided into a training set, a test set and a validation set according to the division ratio; Introducing different machine learning algorithms, using the microalgae characteristic data, bead property data and operation parameter data in the screened multi-dimensional microalgae bead flotation data package as input features and the harvesting efficiency as output feature training rules, and performing model training on the training set in different machine learning algorithms based on the training rules, to obtain the harvesting efficiency prediction model completed by each machine learning algorithm; Establishing an expected recovery efficiency prediction model according to the recovery requirements of the microalgae bead flotation, and determining multiple evaluation indicators of the expected recovery efficiency prediction performance and an ideal threshold value of each evaluation indicator according to the expected recovery efficiency prediction model; The recovery efficiency prediction model trained by each machine learning algorithm is evaluated through multiple evaluation indicators of expected recovery efficiency prediction performance, and a grid search model is introduced simultaneously; Using a grid search model to search for optimal hyperparameters of the recovery efficiency prediction model trained corresponding to each machine learning algorithm, and evaluating based on multiple evaluation indicators and corresponding idealized thresholds, to obtain optimized evaluation indicator values ​​of each recovery efficiency prediction model, and fitting and analyzing the optimized evaluation indicator values ​​through a radar map to select the optimal model, to obtain the ideal recovery efficiency prediction model trained; The grid search model is used to search for the optimal hyperparameters of the recovery efficiency prediction model trained corresponding to each machine learning algorithm and evaluate based on multiple evaluation indicators and corresponding idealized thresholds to obtain the optimized evaluation indicator value of each recovery efficiency prediction model, and the optimized evaluation indicator value is analyzed by radar map fitting and the optimal model is screened to obtain the trained ideal recovery efficiency prediction model, which specifically includes the following steps: During the evaluation process, the grid search model is used to search for the optimal hyperparameters of the recovery efficiency prediction model trained for each machine learning algorithm, so as to improve the recovery efficiency prediction capability of the optimized machine learning algorithm model according to the optimal hyperparameters. At this time, the optimized evaluation index values ​​of each evaluation index output by each trained recovery efficiency prediction model are obtained; Constructing a radar map, fitting the idealized thresholds of the evaluation indicators into the radar map, and generating an idealized evaluation indicator radar map; Fitting each of the trained recovery efficiency prediction models to output optimized evaluation index values ​​of each evaluation index in an idealized evaluation index radar map, and obtaining an optimized evaluation index radar map corresponding to each of the trained recovery efficiency prediction models; Obtaining a radar area value of an idealized evaluation index radar map, defined as a first radar area value, and obtaining a radar area value of each optimized evaluation index radar map, defined as a second radar area value, and calculating the area difference between the second radar area value and the first radar area value one by one to obtain a plurality of area difference values; Only the recovery efficiency prediction model corresponding to the trained recovery efficiency prediction model after optimization evaluation index radar map of the minimum area difference is extracted, and the extracted recovery efficiency prediction model after training is marked as the ideal recovery efficiency prediction model.

2. The method for predicting and optimizing the microalgae flotation recovery efficiency based on machine learning according to claim 1, characterized in that: The step S102 specifically includes the following steps: Obtaining a literature database on microalgae bead flotation on the Web of Science platform, and searching for relevant data in microalgae bead flotation literature through the literature database to obtain microalgae characteristic data, bead property data, operation parameter data, and harvesting efficiency data of microalgae bead flotation; Introducing a data unit conversion tool for the microalgae bead flotation experiment, using the data unit conversion tool to uniformly convert the microalgae characteristic data, bead property data, operating parameter data and harvesting efficiency data, and outputting the microalgae characteristic data, bead property data, operating parameter data and harvesting efficiency data after the uniform units; Obtain the harvesting requirements of microalgae bead flotation, build a data value missing identification library according to the harvesting requirements, and use the data value missing identification library to perform numerical identification and fill in the microalgae characteristic data, bead property data, operation parameter data and harvesting efficiency data after unified units to fill the missing values; After the filling is completed, the microalgae characteristic data, the bead property data, the operation parameter data and the harvesting efficiency data after the unified unit are processed by minimum-maximum normalization and a bundling operation is performed to obtain a multi-dimensional microalgae bead flotation data packet; Preset a Spearman correlation coefficient threshold that affects the harvesting efficiency, calculate the Spearman correlation coefficient between each microalgae characteristic data, bead property data, and operation parameter data in the multi-dimensional microalgae bead flotation data packet and the harvesting efficiency, and obtain a plurality of Spearman correlation coefficients; If the Spearman correlation coefficient is less than the Spearman correlation coefficient threshold, the data corresponding to the Spearman correlation coefficient is deleted from the multidimensional microalgae bead flotation data packet; if the Spearman correlation coefficient is greater than the Spearman correlation coefficient threshold, the data corresponding to the Spearman correlation coefficient is retained in the multidimensional microalgae bead flotation data packet, and finally the screened multidimensional microalgae bead flotation data packet is obtained.

3. The method for predicting and optimizing the microalgae flotation recovery efficiency based on machine learning according to claim 1, characterized in that: The step S106 specifically includes the following steps: Obtaining predetermined microalgae characteristic data, predetermined float property data, and predetermined operation parameter data of microalgae float flotation, and bundling and marking them as predetermined input features; The predetermined input characteristics are estimated and predicted by an ideal recovery efficiency prediction model to obtain a predetermined recovery efficiency value of microalgae bead flotation; Based on the trained ideal recovery efficiency prediction model, a SHAP library is introduced into the big data network, and the SHAP between the predetermined bead properties and the predetermined operating conditions and the predetermined recovery efficiency value is calculated through the SHAP library to obtain multiple SHAP values; Construct a SHAP Summary Plot graph based on the SHAP Summary Plot function, write the multiple SHAP values ​​into the SHAPSummary Plot graph for fitting, and obtain a SHAP Summary Plot visualization graph; The distribution of multiple SHAP values ​​representing the relationship between the given bead properties and the given operating conditions and the given recovery efficiency value is extracted through the SHAP Summary Plot visualization diagram to obtain a SHAP value distribution pattern diagram. The degree of influence of each given bead property and given operating condition on the given recovery efficiency is determined based on the multiple SHAP value distribution patterns of the SHAP value distribution pattern diagram.

4. The method for predicting and optimizing the microalgae flotation recovery efficiency based on machine learning according to claim 1, characterized in that: The step S108 specifically includes the following steps: Arrange each predetermined bead property and the predetermined input characteristics corresponding to the predetermined operating conditions for the predetermined recovery efficiency in descending order based on the influence degree to obtain a descending order table of the predetermined input characteristics; Preset the optimal input feature ranking limit, extract one or more predetermined input features above the optimal input feature ranking limit from the predetermined input feature descending sorting table, and mark them as preferred predetermined input features; An exhaustive method is introduced to change the characteristic values ​​of the one or more preferred predetermined input characteristics to obtain the changed preferred predetermined input characteristics, and the one or more changed preferred predetermined input characteristics are predicted by the trained ideal recovery efficiency prediction model to obtain a new recovery efficiency value; Only the bead properties and operating parameter values ​​under the maximum new recovery efficiency value are selected, and the bead properties and operating parameter values ​​under the maximum new recovery efficiency value are set as the optimal bead flotation recovery conditions; Applying the optimal bead flotation recovery conditions to laboratory microalgae bead flotation related experiments to obtain an actual recovery efficiency value generated by the microalgae bead flotation related experiments, calculating the difference between the actual recovery efficiency value and the established recovery efficiency value, and obtaining a recovery efficiency deviation value; If the recovery efficiency deviation value is greater than the preset recovery efficiency deviation value, the data set of microalgae bead flotation is enriched according to the recovery efficiency deviation value, and the prediction performance of the ideal recovery efficiency prediction model is optimized to improve the recovery efficiency.

5. A microalgae bead flotation harvesting efficiency prediction and optimization system based on machine learning, characterized in that: The microalgae bead flotation recovery efficiency prediction and optimization system includes a memory and a processor. The memory stores a microalgae bead flotation recovery efficiency prediction and optimization method program based on machine learning. When the microalgae bead flotation recovery efficiency prediction and optimization method program is executed by the processor, the microalgae bead flotation recovery efficiency prediction and optimization method steps as described in any one of claims 1 to 4 are implemented.

Images